This study investigates demand for domestic water supply from an innovative borehole system using the contingent valuation method (CVM). We further estimate demand for current service of domestic water supply in residences using the hedonic pricing method (HPM). This is achieved through a survey from rural districts of the Greater Accra Region, Ghana. Interval regression and ordinary least squares (OLS) are applied to investigate the determinants of willingness-to-pay (WTP). We find that monthly WTP values are GH¢35.90 (US$11.45) and GH¢17.59 (US$5.61) in the CVM and HPM, respectively. These values constitute approximately 3–6% of household monthly income which is consistent with earlier studies. For policy purposes, the study recommends the adoption of this cost effective technology to help ease the water burden on society.

## Introduction

In the Millennium Development Goals (MDGs), one of the first targets to have been declared met was the Drinking Water Target. However, rural areas across the world especially in developing countries still lag behind in access to clean drinking water. The United Nations Children's Fund/World Health Organization (UNICEF/WHO, 2014) reports that, 97 out of every 100 people from rural areas in developing countries do not have piped-water, with 14% depending on surface water such as rivers, ponds, or lakes. Sub-Saharan Africa (SSA) needs more attention as it has a worse case relative to other developing regions. It is estimated that only 61% of people in SSA have access to improved water supply relative to over 90% in Latin America and the Caribbean, Northern Africa, and large parts of Asia. Indeed, SSA lags behind the other developing regions in terms of development towards water supply targets.

In moving towards meeting MDG 7, which has further been consolidated into Sustainable Development Goal (SDG) 6 [1], groundwater is considered a reliable improved source for domestic use in SSA. MacDonald et al. (2002) provide some insights into groundwater as a reliable improved option especially in low permeability areas in Africa. They argue that groundwater is a ‘well suited’ source for rural water supply in SSA. It possesses some resilience to the impacts of drought and is relatively cheap to develop and maintain. One major challenge is the kind of improved groundwater (borehole and wells) being provided. These sources are generally the traditional manual types, which require a lot of physical strength from water haulers (mainly women and children) to pump and the groundwater is mostly without filters hence quality is sometimes compromised because of environmental conditions. The water is further exposed to contamination from the point of access to the point of usage. It is important to acknowledge that recent evidence indicates that many improved water supplies suffer from poor reliability (Hunter et al., 2009), and that not all improved water is safe (Levisay & Sameth, 2006). In the view of MacDonald et al. (2002), some proportions of trace constituents in groundwater can make it unsafe and can give rise to health problems. Yet such sources can erroneously be described as improved or safe sources of water. This, to some extent, brings doubts as to whether what is described as ‘improved’ by the international community is the same as ‘safe’1.

In Ghana, the Community Water and Sanitation Agency (CWSA) is the national institution responsible for the provision of safe drinking water and related sanitation services to rural communities. Unfortunately, these communities depend primarily on water from traditional borehole systems which can at best be described as ‘improved’ but not ‘safe’. Another key challenge of the CWSA enshrined in the National Water Policy (NWP, 2007) is how to set tariffs to ensure the sustainability of operations as information on consumers' consumption behaviour is unknown to agents in this market.

In order to avoid future uncertainties regarding the supply of safe drinking water to rural dwellers in a more effective and efficient way, this study primarily aims at estimating the household's demand for an innovative borehole system given that piped-water systems are not available. This is a completely new kind of borehole system. It uses a water pump and it is connected to a solar source of energy supply. The pump generates and supplies water through a filtered-pipe into a communal water tank, which supplies the generated water through a second filtered-pipe(s) to surrounding homes. Thus, we use cost effective resources, such as abundant sunshine and groundwater which is properly filtered, for the design of this innovative system. Water supply from this innovative system can be described as safer and cheaper relative to what is currently being offered. It has the advantage of easing the water burden on women and children with the associated consequential benefits. In addition, it is particularly useful in developing countries where water supply infrastructure is a major problem. We propose this innovative system to the rural community and provide information about households' willingness-to-pay (WTP).

The competing independent valuation approaches generally accepted and used in literature for determining the economic value of non-market goods and services are either based on the revealed/indirect approaches (such as the hedonic pricing method (HPM), travel cost method, take-it-or-leave-it (TIOLI) method, etc.) or stated preference/direct approaches (such as the contingent valuation method (CVM) and choice experiment (CE) method) (see Adamowicz et al., 1994). These valuation approaches also provide economic measures of social benefits needed to inform policy direction. These methods have been used by prominent institutions, such as The World Bank, and applied in both developed and developing countries' contexts (see Briscoe et al., 1990; Bateman et al., 1994; Nauges & Whittington, 2010, etc.). Indeed, it goes without saying that these methods are useful in both settings.

In this paper, we use the CVM, through a hypothetical market design, and the HPM, using rental values of housing units, to measure different aspects of water supply in the rural Greater Accra Region (GAR) of Ghana. We use the CVM to estimate the household's marginal WTP for domestic water supply from the proposed innovative borehole system which captures access to safe water supply, whilst the HPM is used to estimate the household's marginal WTP for the current service which also captures access to current improved water supply in residences. Results from both methods suggest that households place higher value on water from the innovative borehole system than the traditional systems. We find that the household's monthly marginal WTP estimates are GH¢35.90 (US$11.45) and GH¢17.59 (US$5.61) in the CVM and HPM, respectively. In line with the MDGs and SDGs, this study provides information to assist policy makers locally and internationally in their decisions for rural water supply in Ghana. This will help evaluate the socio-economic and health potential of the project as well as determine appropriate tariffs for rural communities which can help design socially equitable fiscal policies.

The rest of the paper is structured as follows. Section 2 presents the empirical review of literature. Section 3 describes the survey design used in carrying out the research. Section 4 presents results and discussions while Section 5 concludes with relevant policy recommendations.

## Empirical literature

In this section, we review studies on WTP which relate to introducing innovative or new products in developing countries and shed light on how studies have been empirically conducted using non-market based valuation methods.

Brouwer et al. (2015) assessed urban and rural demand for a gravity-driven membrane filter for improved drinking-water supply in Kenya. This was a new technology that had not yet been introduced to the market. Respondents had knowledge about other filters and their associated benefits. However, this technology with its benefits were altogether new to respondents. The new technology was based on an extensively tested ultra-low-pressure filtration and flux stabilisation technique. This by design does not require filter cleaning, yet it produces a sufficient amount of water to meet 10–40 L day−1 as required by the WHO and reduces diarrhoea occurrence among children to a maximum of once a year per child. The study combined two stated preference methods namely the CE and the CVM and found the latter to produce conservative and statistically more efficient estimates. The study found that respondents value the new technology positively relative to their current situation. The marginal WTP values in absolute terms were observed to be consistently higher in urban areas than rural areas because of income effect. They concluded that a differentiated marketing strategy is key to a successful introduction of the product in Kenya.

Berry et al. (2012) sought to estimate the WTP for a new product (Kosim filter, a ceramic water filter) introduced to some selected villages in Northern Ghana. This product was not totally new to the entire region as it had been introduced and sold by Pure Home Water2 to some areas but not the areas under study. The respondents were randomly assigned to be offered a water filter applying either the Becker–DeGroot–Marschak (BDM)3 or TIOLI offer. This represented a more typical market transaction because prior to the original survey, demonstrations were made and respondents were further educated about the health benefits of the new product. The respondents saw how the new product worked, tasted the water generated from the new product and asked questions. They were given two weeks to discuss WTP for the new product with their families before participating in the original experiment. The study found evidence that respondents were generally willing to pay for the new product. In addition, they found strong evidence that the WTP implied by the TIOLI was consistently greater than the BDM mechanism. This was justified on two accounts. First, respondents felt they could influence the future price by bidding low. Second, the TIOLI may anchor respondents to higher valuation bids.

In addition, the absence of bathrooms with flush toilets and its health consequences in rural communities and the need for such new facilities within Northern Vietnam motivated Van Minh et al. (2013) to assess WTP for improved sanitation. The economic valuation technique employed was the CVM. Responses were elicited through the iterative bidding game format which involved two stages. First, a sequence of dichotomous choice questions, second, final open-ended questions. The sample size used was 370 households. The units of analysis for the survey were people not having toilets in their residences as of the time of the interview, and were primary income earners as well as decision makers of their respective households. The hypothetical market used comprised descriptions of the good in question (bathroom with a flush toilet and possible benefits). The study found that about two-thirds were willing to pay for an improvement in their current sanitation situation. The economic status of respondents (poor or non-poor) and health knowledge of respondents were the principal influential factors of respondents' WTP.

Another developing country study by Clasen et al. (2004) investigated household demand for water filters with the purpose of reducing diarrhoea in Bolivia. In a six-month trial, water filters were distributed randomly to half of the 50 participating households in the community. The respondents were categorised into a controlled group and intervention group. The respondents generally used customary practices for collecting, storing, and drawing drinking water. Half of the respondents were given filters at the inception of the study, and the other half six months later. Information on WTP was elicited by means of a questionnaire, and they obtained a sample of the pre-intervention drinking water for their baseline data analysis. Participants were randomly allocated by lottery. Half allocated to an intervention group and half allocated to a control group. The study used the CVM to assess WTP for the intervention. The mean response for the maximum amount participants would pay for the filter, was equivalent to US$9.25. Most WTP studies on introducing a new product, have generally followed field experiments and/or hypothetical survey (CVM) methods. However, to the best of our knowledge there is only one study by North & Griffin (1993), which used the HPM to estimate WTP for rural water supply. These authors further confirmed the paucity of studies in this area by indicating that HPM has not yet been applied to WTP for water sources by rural households. The main contribution of our study to this literature is that, it is, to the best of our knowledge, the first study that has applied both CVM and HPM to water supply for rural households. Also, as demonstrated by the various authors in this literature, the relevance of proper description of an innovative product is very critical when dealing with non-market goods. To this end, we used both pictorial and oral approaches for proper description of our innovative borehole system. We observed from our fieldwork that combining both pictorial and oral approaches gave better understanding to respondents. This and other methodological issues are presented in the next section. ## Survey design Household survey data from all the seven districts in rural areas of the GAR were used in this study. We used household responses for the CVM, and housing attributes from the same survey data for the HPM. The total population and number of households in the rural areas of the GAR as reported by the 2010 Population census are 379,099 and 86,090 respectively. We used the Yamane (1967) 4 sample size approach to compute the sample size. We oversampled this to 610 households for higher representation of the population. One response was dropped due to significant missing responses, hence a sample size of 609 households is used in this study. Standard non-market valuation requires that relevant sampling issues (such as technique and sample size) are properly addressed. It is widely known that an inappropriate sampling technique could lead to biased estimates. However, with the unplanned settlements in rural GAR, a multistage quota sampling technique was applied (see Whittington, 1998). This was achieved by clustering the region into seven districts, then into communities. We listed these communities in each district following the Town and Country Planning list of communities and randomly selected the households from these communities within the districts of the region. We sampled one in every two houses. According to our quota, we interviewed all households in the sample houses within the randomly selected communities in the districts. In sum, we applied the multistage quota probability sampling technique in drawing our sample of 610 from the population. A large fraction of rural households in the GAR reside in compound houses together with other households. Communal living effects (where resources are shared) in such compound houses cannot be completely ruled out yet, individual household decisions are mostly the responsibilities of the respective household heads. We therefore considered the entire household as a sampling unit and interviewed whoever the household considered as the household head or decision maker. By the Ghana Statistical Service (2012) definition, the household head is one who is economically and socially responsible for the entire household. Those analysed were household heads who lived in the district, were 18 years old and above, and of sound mind. They should have worked within the last five years and were employed at the time of the survey. However, we also allowed in our sample those who had not worked within the last seven days of the month of the interview. All potential respondents reserved the right to either accept to participate or decline participation. The questionnaire was designed based on two standard national survey questionnaires from Ghana and the United Kingdom. This was subsequently reviewed by survey experts, economists and legal practitioners. The questionnaire was pre-tested in a pilot survey on two different occasions during in-person or face-to-face interviews. This made it necessary for additional amendments to be made to the questionnaire to suit what was practically feasible during this period. The final version of the questionnaire after amendment can be categorised into six sections. For brevity, we summarise them under three main headings: personal data of respondent which comprises all socio-economic and demographic questions; general water, sanitation and environmental questions which includes sources of water supply, water use and reliability, waste disposal forms, and their general knowledge about local and international environmental issues relating to water supply; and environmental valuation questions which consists of the various market designs and WTP questions. The questionnaire was administered by 25 fieldworkers during April–May 2014, which also includes the training of interviewers and coordinators, pilot survey and data entry. ### Valuation approaches and econometric models applied #### The CVM According to Portney (1994, p. 1), ‘[t]he contingent valuation method involves the use of sample surveys (questionnaires) to elicit the willingness of respondents to pay for (generally) hypothetical projects or programs’. The first CVM survey was designed and implemented by Davis (1963). Unlike the revealed preference methods, the CVM has the advantage of capturing both use and non-use values. The CVM follows conventional consumer demand theory, which has it that the quantity demanded of a good is a negative function of price, all else being equal. Respondents in our survey, were asked to place value on the innovative borehole system by answering WTP questions. We define respondent's value in line with a standard household utility function which is convenient with cross-sectional data. Following Whitehead & Blomquist (2005), we specify a standard consumer's utility maximisation function subject to income and prices as: 1 where y denotes the income of the respondent, p and q are the marginal price and quantity of water from the traditional borehole system respectively, and z is a composite of all other goods and services. The solution to the maximisation problem in Equation (1) leads to the indirect utility function, . Alternatively, the minimisation of a consumer's budget, given their utility constraints is shown in Equation (2). 2 Similarly, solution of the problem in Equation (2) yields the consumer's expenditure function, . This can be inverted to obtain the indirect utility function by recognising that and . We demonstrate the entire impact upon a household's welfare by the Hicksian compensating surplus, which essentially shows the amount of income that an individual would be willing to pay for water from the innovative borehole system and, as a result, continue receiving the level of utility (u0) received before the changes. Now the change in the borehole system by introducing the innovative borehole system should be seen as an increment in the consumer's expenditure, . Indeed, WTP for the increment arises and this is shown in Equation (3). 3 We also obtain the compensating surplus function where WTP is a function of some factors, 4 Equation (4) (compensating surplus function) represents a measure of WTP for the innovative borehole system as a function of quantity of water from the innovative system and income of households. Thus, it shows how much each household is willing to sacrifice and yet remain on the same utility level (u0) before the change. For empirical purposes we rewrite the structural economic function given by Equation (4) into an econometric function. Here we assume that the WTP function in Equation (4) takes the following parametric linear form: 5 We rewrite Equation (5) assuming that the maximum amount household i is willing to pay for water from the innovative borehole system is posited as . The error term is represented as which follows a normal distribution function with mean zero and standard deviation (σ). In addition to the regressors in Equation (5), factors such as gender, marital status, and household decision type of respondent have the potential to explain the household's WTP for safe/improved water. Furthermore, these factors are more likely to correlate with income and quantity hence omitting them from the model is likely to lead to omitted variable bias. To ensure consistency and efficiency of the parameters in the WTP function, we account for these additional factors in our empirical specification. We specify our explicit linear functional relationship as: 6 where X is a vector of household characteristics and is a vector of parameters to be estimated. All other variables are as already defined. Hypothetical market description. One essential requirement of CVM studies as outlined by the National Oceanic and Atmospheric Administration (NOAA) is a clear description of the hypothetical market. We describe the innovative borehole system which provides the target commodity as follows. Hypothetical market scenario: • Stage 1 I would want to find out from you, if you value the provision of an improved water supply system in Ghana particularly the rural part of the Greater Accra Region. By improvement it means you are connected to an uninterrupted supply of safe and sufficient water. We have designed an innovative/modernised borehole that is not manual but powered by solar energy so you do not have to pay electricity bills for water generation. This borehole water is filtered, piped and connected directly to your residence. Thus, water flows directly into your residence at all times, the quality is up to acceptable national standards. Generally, we know that every good thing comes at a cost. You may be required to pay a permanent amount that will be factored into your water bills to be provided by the CWSA. • Stage 2 (Refer to pictorial description for further understanding of oral/written description) In the second stage, a picture representing the scenario described in the first stage was shown and narrated to the respondent (see Figure 1). This is also a preferred approach to just describing a hypothetical market (see Whittington & Pagiola, 2012). Fig. 1. Pictorial description of innovative/modernised borehole system. Fig. 1. Pictorial description of innovative/modernised borehole system. In this regard, the two stages were put together and the question asked for the double bound dichotomous choice game was: ‘Suppose you are supplied with this innovative/modernised borehole system as orally and pictorially described, how much would you be willing to pay to fetch a 34 cm bucket of water from this improved system?’ Double bound approach. The double bound design approach is used in this study. According to Whitehead (2000, p. 2), ‘Estimation of the double-bounded willingness to pay data with the interval data econometric model improves the statistical efficiency of WTP estimates relative to single bound models.’ However, this approach is prone to starting point and anchoring effect biases. To correct such biases, Bateman et al. (2002) have suggested the use of a randomised card sorting procedure. In this study, we used a randomised questionnaire sorting (RQS) procedure which in principle is very similar in approach to the card sorting method. In a nutshell, this study used the dichotomous choice double-bound format with RQS. Respondents' bids. We determined marginal WTP through the maximum amount respondents were willing to pay for safe water from the innovative borehole system. The double bound dichotomous choice format used in this study provides three options: yes or no response data, interval data and the maximum amount respondents have stated as their WTP for the good in question. Respondents' responses from the WTP question are used as the dependent variable using different model specifications. The ordinary least squares (OLS) uses the final bid amount stated by the respondent. In the case of the interval regression there were four permutations in the responses from respondents. The Yes-Yes responses, Yes-No responses, No-Yes responses and No-No responses. This approach is presented in Section 3.1.2 (model 2). #### Econometric models applied The double bound dichotomous choice format provides midpoints and interval WTP information. We use two econometric models, namely OLS and interval regression, as robustness checks. Model 1: Ordinary least squares. In this study, the OLS is applied in both valuation methods namely CVM and HPM. We consider a method in which attention is restricted to the final bid for CVM and monthly rental values for HPM. From a broader perspective, we first consider a multiple regression model, using subscript to index the cross-sectional observations and ‘’ to denote the sample size. We represent the multiple regression with parameters and present it as: 7 From Equation (7), given our variables of interest to represent some population, we represent as the dependent variable for observation i, and, = 1, 2, …… k, are the independent variables. The intercept is and represent the slope parameters in the model. We rewrite Equation (7) in a full matrix notation and define as a row vector. We represent y as the vector of observations and the element of y as . Also, X is denoted as the vector of observations on the explanatory variables. Thus, the row of X consists of the vector . With u denoting the vector of unobservable errors, we rewrite for all n observations as: 8 This model assumes to be distributed normally with mean zero and standard deviation . Thus, it is in conformity with the Gauss-Markov5 assumptions underlying the OLS model. This is estimated in both valuation methods used with different functional forms. Model 2: Interval regression. The interval regression model is presented following the double bound dichotomous choice (DBDC) format of individual's WTP which is generally estimated using maximum likelihood (ML) methods (Cameron & Trivedi, 2005). This is achieved by first assuming that the WTP function has a linear functional form and is represented as: 9 where represents the interval within which the true WTP for individual household i can be found. denotes a vector of explanatory variables and a random term which follows a normal distribution function with mean zero and standard deviation . The DBDC format (see Figure 2) suggests that there should be a starting bid . If the respondent says yes, then a second higher bid is offered. For this Yes-Yes option, the lower limit is treated as the second higher bid and the upper limit as positive infinity . Also, in the case of the Yes-No option, the lower limit is the starting bid and the upper limit is the second higher bid . However, if the respondent says No to the starting bid, then a second lower bid is offered . For this No-No option, the upper limit is the second lower bid and the lower limit is zero or negative infinity . Also, for No-Yes options, the lower limit is the second lower bid and the upper limit is the starting bid (See Carson et al., 2003). Fig 2. Double bound dichotomous choice format. Fig 2. Double bound dichotomous choice format. The follows the following definitions: • Yes-Yes option • Yes-No option • No-Yes option • No-No option Taking the cumulative distribution function (CDF) as F the log likelihood function for the DBDC model is represented in Cameron & Trivedi (2005), and Alberini et al. (1997) as: 10 where and are defined as the upper and lower limits or bounds of the interval around WTP. Equation (10) is explicitly formulated and presented for estimation as: 11 The explicit interval regression model presented in Equation (12), is used as the preferred model because of the following two reasons. First, after controlling for district specific effects and starting point bias, we found no evidence of starting point bias or anchoring effect in our results, unlike the OLS models (see Table 3). Second, the interval regression model relative to the OLS estimated in this study, provides the lowest standard errors which suggest a relatively higher level of precision in our estimates. In addition, the parameters in interval regression can be interpreted the same way as in an OLS regression. The ‘Maximum Likelihood (ML) interval technique in log-linear models is unambiguously more reliable than OLS used on interval midpoints' (Cameron & Huppert, 1989, p. 242). We therefore transform Equation (11) and present it as an interval regression function. 12 where is a continuous variable that denotes log of the lower bound and upper bound of respondent's WTP per month for safer water from the innovative borehole system, is log of household head's take-home monthly income in Ghana cedis, represents age in years of respondent, is a dummy variable representing respondent's gender status, is the household size of respondent, is a dummy variable [1, married and 0, unmarried], is a dummy variable [1, main reliable source of drinking water is improved source and 0, otherwise], is a dummy variable [1, access to reservoir in respondent's residence and 0, otherwise], is a dummy variable [1, existence/access to toilet facility in respondent's residence and 0, otherwise], is a dummy variable [1, household residence has fence and 0, otherwise], and is a categorical variable (All the time = 1, Sometimes = 2, Not at all = 3) representing the extent to which respondents use ecologically friendly products. This is used as a proxy to capture respondent's knowledge of environmental issues. is log of the starting point bid to test for starting point bias or anchoring effect in the model, and the error term . #### The hedonic price method The HPM helps to obtain WTP values through the housing market based on rental values or property sale values and attributes of the property. These attributes are generally presented to include structural characteristics (number of stories, number of rooms, nature of floor space, dwelling age, etc.), neighbourhood amenities (distance to public services, distance to work, etc.), and environmental amenities (air and water quality or proximity to open space) (see Van Den Berg & Nauges, 2012). The HPM was first formalised by Rosen (1974). This method is based on the perfect competition and perfect observability of attributes assumption. This assumption is inapplicable in heterogeneous markets such as the property market. Again, all attributes are assumed unrelated and individually evaluable. For a simple modelling of the property market, we assume that how much a household is willing to pay in rental values is conditional on the attributes such as improved source of water in the property. The heterogeneous nature of this market is represented by n attributes. This is presented as: 13 We denote as measuring the amount of attribute in the property, Z. The houses in this market are also assumed to be unique intrinsically (e.g. nature of bedrooms, number of bedrooms, number of bathrooms) and extrinsically (e.g. fence or walls, garden, etc.). Estimating marginal willingness to pay for an attribute includes determining implicit prices of attributes associated with the good, summing the implicit prices obtained, and multiplying by the measure of the attribute to yield the market price of the good (see Devicienti et al., 2004). We re-write Z in an explicit form: 14 where S represents a vector of structural (or residential) characteristics (access to water in residence (R), access to toilet in residence (T), access to electricity in residence (E), residence with fence (F), number of bathroom facilities (NBF), number of toilet facilities (NTF)). N denotes a vector of neighbourhood attributes (water as a district major problem (WDP), distance to nearest hotel or guest or rest house, distance to commercial transport station), and Q represents neighbourhood socio-economic characteristics (mean district savings (MDS)). In line with Rosen's model, we represent our Equations (13) and (14) as: 15 where all variables in Equation (15) are as defined. Choumert et al. (2014) argue that simpler functional forms produce more stable parameter estimates, hence this study uses OLS (see model 1 in Section 3.1.2) with log-log functional form. We re-write Equation (15) following an OLS approach in a more explicit form and specify the econometric model for estimation as: 16 Following the two stage processes of the HPM as presented by Choumert et al. (2014) we determine the implicit marginal price of the different attributes from the aggregate price of the property, . The partial derivative of the aggregate price function relative to an attribute yields the implicit marginal price, , herein referred to as the marginal WTP for the attribute i. In the first stage, we obtain the implicit marginal price by regressing the monthly rental values on the various attributes which include access to improved water supply in the residence. In the second stage, we multiply this implicit value by the average house value to yield the marginal WTP for access to improved water supply per month. ## Results and discussion Below, we present the descriptive statistics (see Tables 1 and 4), results and discussion from the CVM (see Table 2) and then the HPM (see Table 5). Also, we attempt to evaluate whether the two competing methods can be compared. Table 1. Descriptive statistics of variables included in the CVM. Variable Type Description Obs. Mean [%] Std. Dev. Min Max Sign Household income (Y) Continuous Household monthly income in Ghana Cedis (GH¢) 583 591.36 655.85 160 4,400 Age (years) Continuous Respondent's age in years 609 39.31 11.23 21 67 +/ − Male Dummy Gender status of respondent 609 0.52 0.50 +/ − Household size (HH) Continuous Household size of respondent 609 4.54 2.28 17 Marital status (MS) Dummy Marital status of respondent 609 0.63 0.48 Main source (MSD) Dummy Respondent's main reliable source of drinking water is improved source 553 0.11 0.32 Reservoir (R) Dummy Access to water/reservoir (borehole or well etc.) in residence 609 0.42 0.49 Toilet access (T) Dummy Access to toilet facility in residence 609 0.59 0.49 Fence access (F) Dummy Access to fence in residence 609 0.23 0.42 Eco product (Eco) Categorical Use of ecologically friendly products 609 n/a n/a -All the time 229 [37.60%] -Sometimes 326 [53.53%] -Not at all 54 [8.87%] Start bid Discrete Starting point bid 609 25.03 [25% appx]a 11.24 10 40 Lower limit Continuous Lower WTP 609 22.09 11.33 −5 50 n/a Upper limit Continuous Upper WTP 609 42.40 18.90 10 110 n/a Variable Type Description Obs. Mean [%] Std. Dev. Min Max Sign Household income (Y) Continuous Household monthly income in Ghana Cedis (GH¢) 583 591.36 655.85 160 4,400 Age (years) Continuous Respondent's age in years 609 39.31 11.23 21 67 +/ − Male Dummy Gender status of respondent 609 0.52 0.50 +/ − Household size (HH) Continuous Household size of respondent 609 4.54 2.28 17 Marital status (MS) Dummy Marital status of respondent 609 0.63 0.48 Main source (MSD) Dummy Respondent's main reliable source of drinking water is improved source 553 0.11 0.32 Reservoir (R) Dummy Access to water/reservoir (borehole or well etc.) in residence 609 0.42 0.49 Toilet access (T) Dummy Access to toilet facility in residence 609 0.59 0.49 Fence access (F) Dummy Access to fence in residence 609 0.23 0.42 Eco product (Eco) Categorical Use of ecologically friendly products 609 n/a n/a -All the time 229 [37.60%] -Sometimes 326 [53.53%] -Not at all 54 [8.87%] Start bid Discrete Starting point bid 609 25.03 [25% appx]a 11.24 10 40 Lower limit Continuous Lower WTP 609 22.09 11.33 −5 50 n/a Upper limit Continuous Upper WTP 609 42.40 18.90 10 110 n/a Mean and Std. Deviation are rounded off to two decimal places. Not applicable (n/a). [ ] square brackets mean figures are reported in percentages. aPercentage for each of the four bids. Table 2. CVM Results [with (Yes) and without (No) localisation]. (1) (2) (3) (4) Variables OLS (Log-Log) Interval (Log) OLS (Log-Log) Interval (Log) Household income (Log) 0.2135*** 0.1211*** 0.2063*** 0.1149*** (0.030) (0.020) (0.031) (0.021) Age in years −0.0031 −0.0038** −0.0030 −0.0036** (0.002) (0.002) (0.002) (0.002) Male dummy 0.1045** 0.0587* 0.1313*** 0.0770** (0.047) (0.032) (0.047) (0.032) Household size 0.0099 0.0111* 0.0072 0.0103* (0.009) (0.006) (0.009) (0.006) Marital status dummy 0.0592 0.0192 0.0685 0.0224 (0.048) (0.035) (0.047) (0.035) Main source of drinking water 0.4732*** 0.2270*** 0.4843*** 0.2360*** (0.079) (0.053) (0.075) (0.049) Reservoir in residence dummy 0.0539 0.0171 0.0501 0.0191 (0.048) (0.033) (0.047) (0.033) Access to toilet in residence dummy 0.0122 0.0796** 0.0093 0.0769** (0.046) (0.033) (0.048) (0.033) Residence fence-access dummy −0.0549 −0.0248 −0.0628 −0.0319 (0.059) (0.039) (0.060) (0.038) Use of eco-product = 2, Sometimes −0.0451 −0.0161 −0.0642 −0.0252 (0.051) (0.036) (0.051) (0.036) Use of eco-product = 3, Not at all −0.1425* −0.1105* −0.1397* −0.1057* (0.075) (0.057) (0.073) (0.058) Starting point amount (Log) 0.0888** 0.0402 0.0906** 0.0441 (0.044) (0.029) (0.043) (0.029) Constant 1.6679*** 2.5354*** 1.7571*** 2.5898*** (0.220) (0.159) (0.215) (0.156) District dummies No No Yes Yes Observations 521 529 521 529 R-squared [Pseudo R-squared] 0.254 [ 0.11] 0.294 [ 0.12] LR chi (12 & 18 respectively) 121.11*** 136.27*** Mean variance inflation factor (VIF = 1/1- R-squared)a 1.34 1.12 1.42 1.14 (1) (2) (3) (4) Variables OLS (Log-Log) Interval (Log) OLS (Log-Log) Interval (Log) Household income (Log) 0.2135*** 0.1211*** 0.2063*** 0.1149*** (0.030) (0.020) (0.031) (0.021) Age in years −0.0031 −0.0038** −0.0030 −0.0036** (0.002) (0.002) (0.002) (0.002) Male dummy 0.1045** 0.0587* 0.1313*** 0.0770** (0.047) (0.032) (0.047) (0.032) Household size 0.0099 0.0111* 0.0072 0.0103* (0.009) (0.006) (0.009) (0.006) Marital status dummy 0.0592 0.0192 0.0685 0.0224 (0.048) (0.035) (0.047) (0.035) Main source of drinking water 0.4732*** 0.2270*** 0.4843*** 0.2360*** (0.079) (0.053) (0.075) (0.049) Reservoir in residence dummy 0.0539 0.0171 0.0501 0.0191 (0.048) (0.033) (0.047) (0.033) Access to toilet in residence dummy 0.0122 0.0796** 0.0093 0.0769** (0.046) (0.033) (0.048) (0.033) Residence fence-access dummy −0.0549 −0.0248 −0.0628 −0.0319 (0.059) (0.039) (0.060) (0.038) Use of eco-product = 2, Sometimes −0.0451 −0.0161 −0.0642 −0.0252 (0.051) (0.036) (0.051) (0.036) Use of eco-product = 3, Not at all −0.1425* −0.1105* −0.1397* −0.1057* (0.075) (0.057) (0.073) (0.058) Starting point amount (Log) 0.0888** 0.0402 0.0906** 0.0441 (0.044) (0.029) (0.043) (0.029) Constant 1.6679*** 2.5354*** 1.7571*** 2.5898*** (0.220) (0.159) (0.215) (0.156) District dummies No No Yes Yes Observations 521 529 521 529 R-squared [Pseudo R-squared] 0.254 [ 0.11] 0.294 [ 0.12] LR chi (12 & 18 respectively) 121.11*** 136.27*** Mean variance inflation factor (VIF = 1/1- R-squared)a 1.34 1.12 1.42 1.14 Dependent variable: WTP. Robust standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. Note: Exchange rate GH¢1 = US$0.319 as at 15/10/2014.

aPseudo R-squared was used for models 2 and 4.

### CVM results and discussion

Here, four different models are estimated for the CVM and all results are presented in Table 2. The dependent variable for the OLS is log of final bid (WTP), and for the interval regression is log of WTP interval (lower and upper limits). The double bound dichotomous choice format provides an interval within a specific range of true WTP. Based on the assumption that a respondent's final bid could either be overstated or understated, the interval regression intuitively will provide more information on the household's WTP relative to the OLS. In addition, as indicated earlier, the interval regression results show no evidence of starting point bias and produced lower standard errors. Therefore, in the CVM, the most preferred model for our study is the interval regression model where the true WTP is assumed to lie within a certain range of monetary values.

In interpreting our results, we ignore the marginal effects, as they do not represent the monetary values associated with WTP, and focus on the estimated regression coefficients. Generally, the estimated models (see Table 2) are observed to provide quite consistent estimation results, especially with respect to signs of the coefficients across all models. The calculated mean variance inflation factor (VIF) values which range from 1.12 to 1.42 provide evidence of the absence of severe multicollinearity in our models. The goodness of fit (LR chi statistic and R-squared/Pseudo R-squared) support our choice of model 4. All variables to be interpreted assume that ‘all else are held constant’.

We start our discussion with variables capturing the demographic characteristics of respondents. To begin with, the variable Household monthly income was found to be positive and highly significant, implying that a one per cent increase in a household's income will increase their WTP for safer water supply from the innovative borehole system by 0.1149%. Thus, household income elasticity is approximately 0.12 with a confidence interval of 0.07 to 0.16. This suggests that the good in question, although a normal good, is definitely a necessity. Age was expected to be positive and significant due to experience with different water sources, water use and associated health consequences. However, we found age to be negative and significant. This implies that a 1-year increase in respondent's age decreases his/her WTP by 0.36%. This suggests that older people are less willing to pay for improved water supply from the innovative borehole system relative to younger respondents. This could be attributed to the free-rider effect or cohort effect on the part of older people who would expect younger people to pay for them to enjoy. Better still, it may suggest that younger people have different expectations regarding their taste and preferences. This can also be explained by the theory of innovation diffusion where some studies have found that earlier adopters of innovation are younger (Rogers, 1995). A negative effect of age on WTP was also found by Carson et al. (2001). The variable Male was positive and significant. It further shows that males are willing to pay 7.7% more than females in adoption of the new technology. Again, household size is found to be positive and marginally significant. Thus, a unit increase in household size, increases WTP for improved water supply from the innovative borehole system by 1.03%. Marital status was found to be positive but insignificant.

Furthermore, the degree of environmental knowledge is generally important in determining WTP for natural/environmental resources. Use of eco-product which is a categorical variable (all the time [reference category], sometimes, and not at all) was introduced to capture the degree of environmental knowledge. One would expect that respondents who are environmentally informed would express a high WTP to access safer water supply from the innovative borehole system due to health concerns. We found that respondents who do not use ecologically friendly products, express approximately 0.11% lower WTP values relative to those who use ecologically friendly products all the time.

We further introduced log of the starting point amount in the model to capture the possible existence of starting point bias or anchoring effect. We found this to be positive but insignificant in our preferred interval model. This implies that this bias is less important in this model, however, it is important in the OLS model. In short, our preferred model is not being influenced by the randomised starting point amounts used. As shown in Table 3, we proceed to determine the marginal WTP for improved water supply from an innovative borehole system per month using the predicted command in Stata 13.

Table 3.

Predicted WTP measures for reliable and sufficient water supply from an IBS.

Measures Max. WTP for a 34 cm bucket of water from an IBS (pesewas) Max. WTP reliable and sufficient water from an IBS (Cedis)/Day aMax. WTP for reliable and sufficient water from an IBS (Cedis)/Month
Mean 29.92 1.20 35.90
[95% CI] [29.44–30.39] [35.33–36.47]
Median 28.88 1.16 33.50
[95% CI] [28.35–29.36] [32.89–34.06]
% of HH income   0.20% of mean 6.07% of mean
[95% CI] [5.97%–6.17%]
Measures Max. WTP for a 34 cm bucket of water from an IBS (pesewas) Max. WTP reliable and sufficient water from an IBS (Cedis)/Day aMax. WTP for reliable and sufficient water from an IBS (Cedis)/Month
Mean 29.92 1.20 35.90
[95% CI] [29.44–30.39] [35.33–36.47]
Median 28.88 1.16 33.50
[95% CI] [28.35–29.36] [32.89–34.06]
% of HH income   0.20% of mean 6.07% of mean
[95% CI] [5.97%–6.17%]

Note: Computation used the mean household (HH) income of 591.36 and a CI of [538.01–644.70].

IBS implies innovative borehole system.

a[0.2992 × 4 (average amount of water required per day per capita)] × 30(days).

Note: 100 pesewas = 1GH¢.

### HPM results and discussion

We now turn to the HPM. We estimate WTP for improved water supply by housing units with current borehole/well system as an attribute using OLS under the assumption that improved water supply has a perfectly inelastic demand in all the districts within the study area.

We first present the summary descriptive statistics of the HPM in Table 4 and the regression results in Table 5. In the latter case, we present four different models. In model 4 (Table 5), the R-squared and the adjusted R-squared are about 32.4% and 30.3% respectively, higher than all the other models. In addition, the mean VIF value of approximately 1.5 for all models shows the absence of severe multicollinearity. We admit that the models are different, nonetheless, apart from controlling for district specific heterogeneous effects, the coefficient of variation and the mean VIF values make model 4 our preferred model.

Table 4.

Descriptive statistics of variables included in the HPM.

Variable Type Description Obs. Mean Std. Dev. Min Max Sign
MDS (Q) Continuous Neighbourhood socio-economic characteristics (MDS) 609 33.10 10.52 10.28 50.92
Transportation (km) Continuous Distance to nearest commercial transport station 576 1.22 5.26 0.005 60
Hotel (km) Continuous Distance to nearest hotel 543 6.13 14.84 0.001 120
No. of toilets (NTF) Continuous NTF in residence 609 0.74 0.73
Bathrooms (NBR) Continuous Number of bathrooms in residence 609 1.36 0.74
District problem (WDP) Dummy Water as district major problem 609 0.79 0.41 −
Reservoir (R) Dummy Access to water/reservoir (borehole or well etc.) in residence 609 0.42 0.49
Toilet access (T) Dummy Access to toilet facility in residence 609 0.59 0.49
Fence access (F) Dummy Access to fence in residence 609 0.23 0.42
Electricity (E) Dummy Access to electricity in residence 609 0.91 0.29
Rent/month Continuous Rental rate per month 609 61.23 42.56 10 200 n/a
Variable Type Description Obs. Mean Std. Dev. Min Max Sign
MDS (Q) Continuous Neighbourhood socio-economic characteristics (MDS) 609 33.10 10.52 10.28 50.92
Transportation (km) Continuous Distance to nearest commercial transport station 576 1.22 5.26 0.005 60
Hotel (km) Continuous Distance to nearest hotel 543 6.13 14.84 0.001 120
No. of toilets (NTF) Continuous NTF in residence 609 0.74 0.73
Bathrooms (NBR) Continuous Number of bathrooms in residence 609 1.36 0.74
District problem (WDP) Dummy Water as district major problem 609 0.79 0.41 −
Reservoir (R) Dummy Access to water/reservoir (borehole or well etc.) in residence 609 0.42 0.49
Toilet access (T) Dummy Access to toilet facility in residence 609 0.59 0.49
Fence access (F) Dummy Access to fence in residence 609 0.23 0.42
Electricity (E) Dummy Access to electricity in residence 609 0.91 0.29
Rent/month Continuous Rental rate per month 609 61.23 42.56 10 200 n/a

Mean and Std. Dev. are rounded off to two decimal places.

Not applicable (n/a).

Table 5.

Hedonic regression results [with (Yes) and without (No) localisation].

(1) (2) (3) (4)
Variables Lnmonth-rent Lnmonth-rent Month-rent Lnmonth-rent
Access to water in residence_dum 0.2390*** 0.2439*** 21.3914*** 0.2525***
(0.048) (0.048) (3.741) (0.047)
Access to toilet in residence_dum 0.1221 0.1356* 4.4465 0.1013
(0.076) (0.077) (6.044) (0.078)
Water as a district major problem_dum −0.1493** −0.1592*** −10.2999** −0.1501***
(0.059) (0.059) (4.465) (0.058)
Access to electricity_dum 0.0778 0.0744 1.8761 0.0919
(0.085) (0.086) (6.042) (0.086)
Residence fence-type_dum 0.2187*** 0.2213*** 17.1129*** 0.2038***
(0.059) (0.059) (4.875) (0.060)
NBF 0.0490 0.0441 4.2062 0.0666**
(0.034) (0.033) (2.649) (0.033)
NTF 0.1356** 0.1398** 9.9502** 0.1315**
(0.056) (0.056) (4.951) (0.059)
Distance to nearest hotel (km) −0.0053*** −0.0055*** −0.4154*** −0.0061***
(0.001) (0.001) (0.095) (0.001)
Distance to transport station (km) −0.0052* −0.0053* −0.4507* −0.0025
(0.003) (0.003) (0.232) (0.003)
Mean_district_savings 0.0072***
(0.002)
Mean_district_savings (Log)  0.1266*
(0.069)
Constant 3.3807*** 3.1924*** 55.3405*** 3.8047***
(0.141) (0.268) (8.103) (0.111)
District dummies No No Yes Yes
Observations 529 529 529 529
R-squared 0.285 0.277 0.286 0.324
Adjusted R-squared 0.271 0.263 0.264 0.303
Mean variance inflation factor (VIF) 1.49 1.49 1.48 1.48
(1) (2) (3) (4)
Variables Lnmonth-rent Lnmonth-rent Month-rent Lnmonth-rent
Access to water in residence_dum 0.2390*** 0.2439*** 21.3914*** 0.2525***
(0.048) (0.048) (3.741) (0.047)
Access to toilet in residence_dum 0.1221 0.1356* 4.4465 0.1013
(0.076) (0.077) (6.044) (0.078)
Water as a district major problem_dum −0.1493** −0.1592*** −10.2999** −0.1501***
(0.059) (0.059) (4.465) (0.058)
Access to electricity_dum 0.0778 0.0744 1.8761 0.0919
(0.085) (0.086) (6.042) (0.086)
Residence fence-type_dum 0.2187*** 0.2213*** 17.1129*** 0.2038***
(0.059) (0.059) (4.875) (0.060)
NBF 0.0490 0.0441 4.2062 0.0666**
(0.034) (0.033) (2.649) (0.033)
NTF 0.1356** 0.1398** 9.9502** 0.1315**
(0.056) (0.056) (4.951) (0.059)
Distance to nearest hotel (km) −0.0053*** −0.0055*** −0.4154*** −0.0061***
(0.001) (0.001) (0.095) (0.001)
Distance to transport station (km) −0.0052* −0.0053* −0.4507* −0.0025
(0.003) (0.003) (0.232) (0.003)
Mean_district_savings 0.0072***
(0.002)
Mean_district_savings (Log)  0.1266*
(0.069)
Constant 3.3807*** 3.1924*** 55.3405*** 3.8047***
(0.141) (0.268) (8.103) (0.111)
District dummies No No Yes Yes
Observations 529 529 529 529
R-squared 0.285 0.277 0.286 0.324
Adjusted R-squared 0.271 0.263 0.264 0.303
Mean variance inflation factor (VIF) 1.49 1.49 1.48 1.48

Dependent variable: Rent per month in Ghana cedis.

Robust standard errors in parentheses.

*** p < 0.01, ** p < 0.05, * p < 0.1.

We also observed that all the explanatory variables had the expected signs. However, except for three variables: Access to toilet in residence, Access to electricity and Distance to transport station (km), all estimated coefficients are found to be statistically significant at various levels of significance. In interpreting our variables, we further assume that ‘all else are held constant’.

Our discussion is presented under water related residential and neighbourhood characteristics, other residential/structural characteristics, and other neighbourhood characteristics. The water related explanatory variables presented in our model include: Reservoir in residence, Access to toilet facility in residence, Water as a district major problem. The other residential/structural characteristics include: Access to electricity, Residence fence, Number of bathrooms, and NTF. Lastly, the other neighbourhood characteristics include: Distance to nearest hotel (km), Distance to transport station (km) and Mean_district_savings.

Next is the residential/structural characteristics. The study finds that all the variables in the preferred model 4 relating to residential/structural characteristics had the expected a priori signs (see Sirmans et al., 2005). First, Access to electricity is found to be positive albeit insignificant. This could be explained by the fact that 79% of rural people are without access to electricity in their homes (ibid). Second, Residence fence was introduced to capture the possible free-riding effect in areas characterised by communal living. This provides a positive and very high statistically significant effect on rental value. That is, fenced residences pay 20.38% more in rental values relative to unfenced residences. Third, Number of bathrooms had the expected positive sign on rental values, although this is seen not to be significant in the other models except the preferred model 4. The result suggests that if the number of bathrooms in a residence increase by one, households will pay 6.66% more in rental values. In addition, Number of toilets is positive and significant in all estimated models. This implies that if the number of toilets in a residence increase by one, households will pay 13.15% more in rental values. In effect, better residential characteristics evidenced by quantity and quality of residential characteristics are seen to increase rental values.

The quality of neighbourhood characteristics is expected to affect rental values. For example, Distance to nearest hotel (km) which captures some degree of prestige, environmental quality, security, affluence, etc., definitely will increase rental values. This variable is seen to provide evidence of a negative and highly statistically significant effect on rental values. This implies that residences that are located within a kilometre range, closer to a hotel, increase rental values by 0.61%. More so, we find the Distance to transport station (km) variable to be negative and significant in all models except in our preferred model.

To further evaluate the potential effect of district wealth heterogeneity on rental values, we introduced the Mean_district_savings variable in models 1 and 2 as a proxy for income and wealth. This could not have been included in models 3 and 4 because of severe collinearity with district dummies. We find evidence of a positive and significant effect of the Mean_district_savings on rental values in both models. It can be inferred that districts with high income and savings (or wealthy households) tend to pay more in rental values. This satisfies the scope sensitivity test commonly found in valuation studies.

We now turn our attention to the computation of the marginal WTP for having access to reliable water supply which is proxied with access to improved water supply in residence. Given that the variable of interest is dummy, we compute the relative change in rental values with results from Table 5 (model 4) using the delta method. This study finds that the average amount households will be prepared to pay per month for access to water in residence is GH¢ 17.59 which constitutes 2.98% and 2.68% of the mean-district-income and mean-household-income per month respectively (see Table 6). According to Bartik (1988) and Choumert et al. (2014), this should be interpreted as an upper bound value because the utility dummy may include unobserved attributes and utilities.

Table 6.

Predicted increase in the value of house with access to water supply.

Marginal implicit house value per month (GH¢) Current average HH expenditure on water per month (GH¢) Increment as % of monthly district-income Increment as % of monthly household income
Meana Mean Meanb Meanc
17.59 41.554 2.98% 2. 68%
[10.34–24.85] [39.41–43.69] [1.75%–4.20%] [1.58%–3.79%]
Marginal implicit house value per month (GH¢) Current average HH expenditure on water per month (GH¢) Increment as % of monthly district-income Increment as % of monthly household income
Meana Mean Meanb Meanc
17.59 41.554 2.98% 2. 68%
[10.34–24.85] [39.41–43.69] [1.75%–4.20%] [1.58%–3.79%]

[] Denote confidence intervals estimated at 95%.

aRelative change (water dummy) × Average house value = 0.28724 × 61.23064 = 17.59 ≈ GH¢18 per month.

bMarginal implicit house value/Average district income = 17.59/591.3551 = 0.0298 × 100 = 2.98%.

cMarginal implicit house value/Average household income = 17.59/655.85 = 0.0268 × 100 = 2.68%.

### WTP estimates: can we directly compare our estimates

The CVM and HPM are valuation methods employed to estimate WTP for improved supply of rural water. However, it needs to be pointed out that in application, they could capture different things yet provide relevant estimates that are worthwhile for policy purposes. The estimated results presented in Tables 3 and 6 are summarised in Table 7. In Table 7, the results are presented in both Ghana Cedis (GH¢) and in United States dollars (US$) for easy understanding. Table 7. A summary of CVM and HPM estimates. Method WTP (GH¢)/M* 95% CI WTP US$/M* % of income index
CVM 35.90 [35.33–36.47] 11.45 6.07%
HPM 17.59 [10.34–24.85] 5.61 2.68%
Method WTP (GH¢)/M* 95% CI WTP US$/M* % of income index CVM 35.90 [35.33–36.47] 11.45 6.07% HPM 17.59 [10.34–24.85] 5.61 2.68% CI denotes confidence interval. *M = month; GH¢ = US$0.319 as at 15/10/2014.

From Table 7, it is important to acknowledge that the CVM used here seeks to measure how much respondents are willing to pay per month for improved and safer water supply from an innovative borehole system. The values captured by this method include use values of an improved system over what is currently being used. The HPM seeks to measure the economic value of improved water supply from an amenity (reservoir, i.e. traditional borehole or well) in residence per month through house prices, or how much households with access to water are willing to pay per month. Stated differently, the HPM provides estimates of the additional amounts households with access to water supply in residence are willing to pay per month in rental values. This captures the use values of the current service only. Therefore, we expected estimates from the CVM to be greater than from the HPM. The results show that CVM estimates are much more precise than HPM ones at 95% confidence interval.

The HPM estimate of GH¢17.59 (US$5.61) per month and the CVM of GH¢35.90 (US$11.45) per month constitute approximately 3–6% of household income. Payment of this by potential beneficiaries represents a sensible trade-off that people might make towards policy implementation (see Carson, 2012). However, it is important to reiterate that these estimates are capturing entirely different things and cannot be directly compared in our case. According to McPhail (1993, p. 1), ‘… most utilities and donors assume that, as long as the cost of potable water to the household falls below 5% of household income, then it is ‘affordable’ and the household will make a connection to the system and be able to pay the subsequent recurrent charges’. Similar assertions have also been made by Whittington et al. (1990) to that effect. In view of this, we may conclude that our estimates are within a reasonable range of affordability and that respondents have shown a positive attitude towards the services.

## Policy implications and conclusion

Towards achieving MDG 7 (now consolidated into SDG 6), this study focuses on providing information on households' WTP for sustainable, safe rural water supply in Ghana. This study is important against the background that unsustainable planning and management have largely been attributed to absence of information on consumer's WTP for water supply services.

Indeed, policy makers are not fully informed about consumer's WTP to have access to their current state of water supply as well as improvement in water supply. This has triggered a myriad of studies in this area with the primary motive of contributing to policies relevant to sustainable safe water supply. To this end, we use the HPM to capture WTP for the current service, and CVM to also capture WTP for improvement in the service through introduction of an innovative borehole system. We therefore provide policy recommendations as follows.

We recommend that to achieve SDG 6(1) of safe and affordable drinking water supply, an innovative and affordable system with relevance to women and children like this should be considered. Alternatively, the Ghana Water Company Limited and CWSA should consider using our estimates for a cost benefit analysis of this project to extend piped water services to the rural areas. Also the estimated WTP may be used to encourage households to adopt such safe appliances across the country and elsewhere.

Currently, the world is full of praise for meeting the access to improved water target, as enshrined in MDG 7. However, we argue, following Hunter et al. (2009) and Levisay & Sameth (2006), that not all improved water supplies are safe. In order to ensure that the current SDG is achieved with safe water for rural Ghana, this study proposes an innovative borehole system and estimates the demand for water from this system. This is achieved by using the HPM to capture the economic value of an existing system (i.e. marginal WTP for access to improved water from the traditional borehole and well) in residences and CVM to capture the same for improved and safer water from the innovative borehole system. Our results suggest that households support the improvement in their water supply and are willing to pay about 3–6% of their income for it.

In short, we present one of the first estimates of the economic values for rural water supply using both the HPM and CVM to capture for current service (improved water) and improvement in current service (safer-water), through a proposed innovative borehole system, in a developing country. These results may be applied to other developing countries with similar characteristics without any loss of generality.

## Acknowledgements

The author acknowledges support from the following institutions: GETFund (Ghana), Central University (Ghana), University of East Anglia (UK). The author wishes to express his heartfelt gratitude to Prof. Peter G. Moffatt, Dr Corrado Di Maria, Dr Silvia Ferrini, Dr Philomena Bacon, Dr Bereket Kebede, Dr George Adu, Franklin Amuakwa-Mensah, anonymous reviewers and the editorial team for their excellent work that has refined the quality of the paper. All errors and omissions are his own.

1

‘Safe drinking water is water with microbial, chemical and physical characteristics that meet WHO guidelines or national standards on drinking water quality’ (WHO, 2015; www.who.int/water_sanitation_health/mdg1/en). In this study safe water is referred to as water supply from a piped system or treated borehole water, etc. and improved water is defined as water supply from boreholes, wells, etc. which is not necessarily treated.

2

A Ghana-based non-profit organisation.

3

BDM is a method for measuring utility by a single-response sequential method. It is considered to be an incentive-compatible procedure used in experimental economics to measure willingness-to-pay.

4

Yamane (1967) sample size determination approach: . Where n is the sample size, N is the size of the population, and e is the error level or level of precision.

5

For simplified discussions of the Gauss-Markov assumptions, see Wooldridge (2014, p. 93, 2006).

## References

References
W.
Louviere
J.
Williams
M.
(
1994
).
Combining revealed and stated preference methods for valuing environmental amenities
.
Journal of Environmental Economics and Management
26
(
3
),
271
292
.
Alberini
A.
Kanninen
B.
Carson
R. T.
(
1997
).
Modeling response incentive effects in dichotomous choice contingent valuation data
.
Land Economics
73
(
3
),
309
324
.
Apt
N.
Amankrah
J.
(
2004
).
Assessing Ghanaian Insecurities at the Household Level. ILO Socio-economic Security Programme: Confronting Economic Insecurity in Africa, edited by Rajendra Paratian and Sukti Dasgupta
,
ILO Office
.
Bartik
T. J.
(
1988
).
Measuring the benefits of amenity improvements
.
Land Economics
64
(
2
),
172
183
.
Bateman
I.
Willis
K.
Garrod
G.
(
1994
).
Consistency between contingent valuation estimates: a comparison of two studies of UK National Parks
.
Journal of the Regional Studies Association
28
(
5
),
457
474
.
Bateman
I. J.
Carson
R. T.
Day
B.
Hanemann
W. M.
Hanley
N.
Hett
T.
Jones-Lee
M.
Loomes
G.
Mourato
S.
Özdemiroglu
E.
Pearce
D. W.
Sugden
R.
Swanson
J.
(
2002
).
Economic Valuation with Stated Preference Techniques: A Manual
.
Edward Elgar Publishing
,
Cheltenham
.
Berry
J.
Fischer
G.
Guiteras
R.
(
2012
).
Eliciting and utilizing willingness to pay: evidence from field trials in Northern Ghana. Unpublished manuscript
.
Briscoe
J.
de Castro
P. F.
Griffin
C.
North
J.
Olsen
O.
(
1990
).
Toward equitable and sustainable rural water supplies: a contingent valuation study in Brazil
.
The World Bank Economic Review
4
(
2
),
115
134
.
Brouwer
R.
Job
F. C.
van der Kroon
B.
Johnston
R.
(
2015
).
Comparing willingness to pay for improved drinking-water quality using stated preference methods in rural and urban Kenya
.
Applied Health Economics and Health Policy
13
(
1
),
81
94
.
Cameron
A. C.
Trivedi
P. K.
(
2005
).
Microeconometrics: Methods and Applications
.
Cambridge University Press
,
New York
,
USA
.
Cameron
T. A.
Huppert
D. D.
(
1989
).
OLS versus ML estimation of non-market resource values with payment card interval data
.
Journal of Environmental Economics and Management
17
(
3
),
230
246
.
Carson
R. T.
(
2012
).
Contingent valuation: a practical alternative when prices aren't available
.
Journal of Economic Perspectives
26
(
4
),
27
42
.
Carson
R. T.
Flores
N. E.
N. F.
(
2001
).
Contingent valuation: controversies and evidence
.
Environmental and Resource Economics
19
(
2
),
173
210
.
Carson
R. T.
Mitchell
R. C.
Hanemann
M.
Kopp
R. J.
Presser
S.
Ruud
P. A.
(
2003
).
Contingent valuation and lost passive use: damages from the Exxon Valdez oil spill
.
Environmental and Resource Economics
25
(
3
),
257
286
.
Choumert
J.
Stage
J.
Uwera
C.
(
2014
).
Access to water as determinant of rental values: a housing hedonic analysis in Rwanda
.
Journal of Housing Economics
26
,
48
54
.
Clasen
T. F.
Brown
J.
Collin
S.
Suntura
O.
Cairncross
S.
(
2004
).
Reducing diarrhea through the use of household-based ceramic water filters: a randomized, controlled trial in rural Bolivia
.
American Journal of Tropical Medicine and Hygiene
70
(
6
),
651
657
.
Davis
R. K.
(
1963
).
The value of outdoor recreation: an economic study of the Maine woods
.
PhD Thesis
,
Harvard University
,
Baltimore
.
Devicienti
F.
Klytchnikova
I.
Paternostro
S.
(
2004
).
Willingness to Pay for Water and Energy: An Introductory Guide to Contingent Valuation and Coping Cost Techniques, Energy Working Notes, Energy and Mining Sector Board, 3
.
Government of Ghana, Ghana Statistical Service
(
2012
).
2010 Population and Housing Census, Summary Report of Final Results
.
Sakoa Press Limited
,
Ghana
.
Hunter
P. R.
Zmirou-Navier
D.
Hartemann
P.
(
2009
).
Estimating the impact on health of poor reliability of drinking water interventions in developing countries
.
Science of the Total Environment
407
(
8
),
2621
2624
.
Levisay
M.
Sameth
C.
(
2006
).
Measuring Rural Water Supply Access: Findings from a Comparative Analysis of Cambodian National Surveys
.
Project Report, Water and Sanitation Program, Ministry of Rural Development
,
Cambodia
.
MacDonald
A. M.
Davies
J.
Dochartaigh
B.
(
2002
).
Simple Methods for Assessing Groundwater Resources in Low Permeability Areas of Africa
.
British Geological Society
,
Nottingham
,
UK
.
McPhail
A. A.
(
1993
).
The ‘five percent rule’ for improved water service: can households afford more?
World Development
21
(
6
),
963
973
.
Nauges
C.
Whittington
D.
(
2010
).
Estimation of water demand in developing countries: An overview
.
The World Bank Research Observer
25
(
2
),
263
294
.
North
J. H.
Griffin
C. C.
(
1993
).
Water source as a housing characteristic: hedonic property valuation and willingness to pay for water
.
Water Resources Research
29
(
7
),
1923
1929
.
NWP, MWRWH-Government of Ghana
(
2007
). .
Portney
P. R.
(
1994
).
The contingent valuation debate: why economists should care
.
Journal of Economic Perspectives
8
(
4
),
3
17
.
Rogers
E. M.
(
1995
).
Diffusion of Innovations
,
4th edn
.
Free Press
,
New York
.
Rosen
S.
(
1974
).
Hedonic prices and implicit markets: product differentiation in pure competition
.
Journal of Political Economy
82
(
1
),
34
55
.
Sirmans
S.
Macpherson
D.
Zietz
E.
(
2005
).
The composition of hedonic pricing models
.
Journal of Real Estate Literature
13
(
1
),
1
44
.
United Nations
(
2006
).
Human Development Report 2006: Beyond Scarcity: Power, Poverty and the Global Water Crisis
.
United Nations Development Programme
,
Van Den Berg
C.
Nauges
C.
(
2012
).
The willingness to pay for access to piped water: a hedonic analysis of house prices in Southwest Sri Lanka
.
Letters in Spatial and Resource Sciences
5
(
3
),
151
166
.
Van Minh
H.
Nguyen-Viet
H.
Thanh
N. H.
Yang
J. C.
(
2013
).
Assessing willingness to pay for improved sanitation in rural Vietnam
.
Environmental Health and Preventive Medicine
18
(
4
),
275
284
.
J.
(
2000
).
Anchoring and Shift in Multiple Bound Contingent Valuation (No. 0004)
.
Department of Economics, East Carolina University Greenville
,
NC 27858
.
J. C.
Blomquist
G. C.
(
2006
).
The use of contingent valuation in benefit-cost analysis
. In:
A. Alberini & J. R. Kahn (eds)
.
Handbook on Contingent Valuation
.
Edward Elgar
,
Northampton
, pp.
92
115
.
Whittington
D.
(
1998
).
Administering contingent valuation surveys in developing countries
.
World Development
26
(
1
),
21
30
.
Whittington
D.
Pagiola
S.
(
2012
).
Using Contingent Valuation in the Design of Payments for Environmental Services Mechanisms: A Review and Assessment
.
The World Bank Observer, Oxford University Press
,
UK
.
Whittington
D.
Briscoe
J.
Mu
X.
Barron
W.
(
1990
).
Estimating the willingness to pay for water services in developing countries: a case study of the use of contingent valuation surveys in Southern Haiti
.
Economic Development and Cultural Change
38
(
2
),
293
311
.
WHO & UNICEF
(
2014
).
Progress on Drinking Water and Sanitation
.
WHO Press, World Health Organization
,
Geneva
,
Switzerland
.
WHO
(
2015
).
Health through safe drinking water and basic sanitation. Available from: http://www.who.int/water_sanitation_health/mdg1/en, accessed 31 August 2015
.
Wooldridge
J.
(
2014
).
Introduction to Econometrics
.
Cengage Learning EMEA
,
London
.
Yamane
T.
(
1967
).
Statistics, an Introductory Analysis
,
2nd edn
.
Harper and Row
,
New York
.