Estimating averting expenditure in domestic water use: evidence from Ghana

Safe water is described as an important resource for the survival of mankind. The outbreak of the Covid-19 pandemic has made safe water ‘super’ important and critical for the survival of mankind. Most developing countries, especially in Africa, incur additional costs in order to enjoy improved, if not safe domestic water supply. Using the averting expenditure method, this study estimates how much urban households in the Greater Accra Region of Ghana spend to improve the quality of domestic water they use. The study provides evidence that households spend Ghs84.30 ($14.70) per month, which constitutes 13.25% of their income. These estimates are very informative to the supplier in determining the economic viability of making the required quality of water available to households.


INTRODUCTION
Access to potable water and residential water supplies in Ghana and other developing countries has been increasing steadily. These success stories follow the concerted global efforts in driving goals such as the Millennium Development Goals (MDGs) and the Sustainable Development Goals (SDGs) in addition to national-level policies. In Ghana, the gains include, but are not limited to, meeting the drinking water target under the MDGs a decade before the deadline, continuous increase in water coverage, reduction in water-related diseases and its associated impact on infant and child mortality rates (Monney & Antwi-Agyei approach by most Ghanaian water users is to adopt averting behaviour as an improvement mechanism to domestic water before use. However, this approach involves time and effort, cost of averting materials or technique, relatively higher water cost, etc. Regardless of the averting behaviour, water quality for domestic use remains questionable. Currently, urban areas in Ghana experience inadequate supply of domestic water because of rising population levels and the high cost incurred by the Ghana Water Company Limited (GWCL) in improving the polluted sources of surface water. This gloomy situation is worse for rural communities where there is a lack of piped-water infrastructure layout. In the few rural communities with piped-water supply, water hardly flows, hence their reliance on boreholes provided mainly by the Community Water and Sanitation Agency (Amoah ). The need for stakeholders to act now follows the uncertainty that surrounds the supply of safe domestic water and associated health-related impact on human capital now and in the future. Safe drinking water is water with microbial, chemical and physical characteristics that meet WHO guidelines or national standards on drinking water quality (WHO ; www.who.int/ water_sanitation_health/mdg1/en).
The primary question this study seeks to answer is, can the averting expenditure be estimated to determine the willingness-to-pay (WTP) for domestic pipe-borne water supply in residences? One of the revealed preferred methods that has proven reliable for the estimation of WTP for water supply is the averting behaviour method (ABM) (see Van Houtven et al. ). In this study, we define ABM as the actual practice(s) undertaken to prevent the harmful effect of water contamination. The associated cost incurred for engaging in these practices is the averting cost or expenditure. Unfortunately, there is a paucity of studies on this method in developing countries (Pattanayak et al. ).
To the best of our knowledge, no study has provided water-related averting behaviour estimates in Ghana. The authors concluded that averting behaviour is essential to water use in Hawassa, Ethiopia.
We infer that because the degree of water contamination or water quality is not the same across countries and communities (be it developed or developing), averting behaviour and associated cost estimates can also not be the same. The pieces of evidence from the previous studies reviewed, and the findings of the current study are externally validated by Amoah & Moffatt (), who have shown with empirical literature that households generally are willing to spend an average of 2%-18% of their income to improve the quality of their domestic water.
The main contribution of the study is using ABM and a survey approach to determine how much it costs urban households in the Greater Accra Region (GAR) to improve the quality of water before use. In addition, it provides a demand-side estimate to inform private sector decisions for water sector investment. Also, this study contributes to the empirical literature on averting behaviour and domestic water use in response to the paucity of literature on the application of ABM in developing countries, as acknowledged by Pattanayak et al. ().

Urban water situation in Ghana
Ghana is a fairly water-endowed country. The Ghana Water Company Limited (GWCL) is a state-owned company responsible for urban water supply which is almost entirely from surface water resources. The GWCL, therefore, has abundant surface water resources to meet the current and future urban consumptive demand (Yirenkyi-Fianko ).
Unfortunately, the quality of surface water and groundwater sources keep deteriorating at an alarming rate. This is mainly attributed to rapid population growth without a cor- shows that sachet water is the main source of household drinking water while the pipe-borne water is for general use (washing, bathing, cleaning, etc.).

Data collection process
This study focuses on the ten districts in the GAR as defined by the 2010 Housing and Population Census by the GSS (). The Region has a total population and number of households of 4,010,054 and 1,036,426, respectively. Out of the latter, the urban and rural household population constitutes 766,955 and 269,471, respectively. Given that the rural case is dire and piped infrastructure is lacking, we focused on the urban area, hence our reference sampling frame is 733,955 households. In anticipation of a representative sample, the Yamane () sample size formula was used to compute the sample size which yielded approximately 400 households. The study over-sampled to as high as 1,650 households.
The instrument used for the data collection was a structured questionnaire. The first section of the questionnaire had the demographic or socio-economic data of the respondent (personal data), the other sections had water, sanitation and other environmentally related questions.
This included improved water valuation questions such as averting behaviour and expenditure.
The questionnaire was administered by 20 trained fieldworkers who were supervised by four coordinators. The principal investigator was responsible for the overarching supervision and ethical adherence by the entire team.
Before the commencement of the main survey, the team undertook a pre-pilot survey before the actual pilot survey. This was done to avoid possible trial and error before the main survey. All responses in both pilot surveys were analysed to inform possible amendments of some of the questions before the survey. It is important to point out that the pilot datasets were not included in the final dataset.
Given the unplanned nature of settlements in most urban areas in Africa, of which the GAR is no exception, we applied a multistage probability sampling technique. This was achieved by first defining the various districts as unique clusters, followed by listing the communities in each cluster as prescribed by the Town and Country Planning Department. Households within each community were also listed and randomly selected until our expected quota was reached. In all, based on a multistage probability sampling technique, this study relies on 1,648 observations after two observations were excluded due to missing data points. This gives a very high response rate of about 99% which is very common in most surveys in the GAR (e.g.,

Econometric modelling
The principal aim of modelling the averting expenditure is to calculate the conditional averting average cost (conditional mean) instead of just the mean of averting cost (unconditional). Second, this study seeks to determine the drivers of averting behaviour and provide evidence that supports the internal and external validity of the study. The data show that approximately 10% of the sample do not incur any averting cost as they enjoy regular flow of safe piped water that does not require an averting behaviour. In the regression and the averting cost estimates, the study excluded the 10% to only focus on households that engage in averting behaviour and have incurred an expenditure in that respect. Thus, the discrete dependent variable, total averting expenditure or cost per household (C i ), depends on socio-economic, demographic, context relevant variables and district dummies. Following Amoah et al. (), the econometric model is specified as: where C i is as already defined, X i is a vector of independent variables used for the estimation, α and β are the unknown parameters to be estimated. The stochastic term, u i , follows a standard normal distribution. In order to interpret the results as elasticities, we transformed the model into a natural logarithm form. The raw data and their descriptive statistics are presented in Table 1 before transformations.
From Table 1, the averting cost per day reported ranges from a minimum of Ghs0.06 (6 pesewas or 1 cent) and a maximum of Ghs36.48 ($6.29). The unconditional mean averting expenditure or cost is Ghs4.36 ($0.75) per day.
The average household income per month is Ghs636.37 ($109.72) which is quite close to the national estimate of Ghs544 ($93.79) (GSS ). The average age of respondents is 39 years with 89% of household heads being males. The majority (59%) of the respondents are married.
Sixty-eight per cent claim they have reliable or regular supply, yet they still engage in averting behaviour. This shows that the quality of water is not always as perfect or safe as one would expect. The household life cycle variable shows the various stages of households with the average representation being families with or without children.

RESULTS AND DISCUSSION
In Table 2, we show the various sources of water being used by urban households in the GAR. The main sources are categorized into water for drinking and water for general use. For the drinking water sources, it is observed that approximately 82% of the respondents use sachet or bottled (packaged) water. This is due to unreliability and poor quality from the pipe-borne system. Hence, most households depend on the commercial (public or private) standpipe water sources which are relatively reliable because of available storage facilities. Consistent with our evidence, the Ghana Statistical Service (GSS ) reports that in the urban areas of Ghana, sachet water constitutes the main source of household drinking water and where households use pipe-borne water, they mostly rely on either private or public standpipes. In the case of water for general use (washing, bathing, cleaning, etc.), most households mainly depend on pipeborne related sources. Given that pipe-borne sources do not flow reliably, most households use water-saving containers to reserve water during piped-water opening or flowing days.
Such sources are susceptible to contamination, hence the need for averting practices. Evidently, the GSS () reports that households in urban areas mainly use piped-system and/or tanker services (pipe-borne or borehole or wells).
Overall, our results suggest that most households in the urban areas rely mainly on packaged water for drinking while pipe-borne sources are used for other general household purposes. This evidence may be driven by pipe-borne water quality concerns in the GAR.
Next, we show how most households improve their quality of water before use. In urban GAR of Ghana, the estimated total averting cost for improved water supply is deemed very informative to policy-makers as well as the supplier (GWCL). The total averting cost provides an estimate of the additional  amount households would be willing to pay for improvement in their domestic water supply to avoid averting behaviour. Next, we investigate the drivers of averting cost.

Regression analysis of drivers of averting cost
To commence our discussion, we perform some diagnostic tests to ensure the validity of our results. In cross-sectional data analysis, multicollinearity and heteroskedasticity tests are a pre-requisite. To ensure that no exact linear relationship exists among the independent variables, we used the pairwise correlation matrix to compute their respective degree of correlation. From the test results (Table A1 in the Appendix), the highest degree of correlation is found between age and marital status with a correlation coefficient of approximately 0.29. It suggests that there is no concern for multicollinearity among the variables in the data. The other problem is heteroscedasticity, which arises when there is unequal variance in the error term. To control for this in our study, the coefficients are estimated with White's heteroskedasticity-consistent robust standard errors. In addition, we do not assume homogeneity among the districts, so we introduced district fixed effects to account for possible bias that may arise as a result of differences in the districts. We report an R-squared (coefficient of variation) of 15% which fits the minimum value for reliable valuation studies as proposed by Mitchel & Carson ().
The F-statistics of 8.73 and a p-value of 0.00 provide evidence that the overall model is fit for prediction.
Following the evidence from the diagnostic tests, we confidently argue that the estimated results are not severely biased and can be trusted in making policy decisions. The interpretation and discussion of our results assume that all other factors are held constant (ceteris paribus).
In line with the theoretical construct which satisfies an internal validity test, we argue that a consumer's demand is expected to vary with his/her income. If the good is a normal good, we expect that an increase in income will increase consumption of the good in question. In Table 5, we found that there is a positive and statistically significant relationship between income and averting expenditure. That is, if a household's income increases by 1%, the average averting expenditure will increase by 0.0880%. This suggests that water is not just a normal good but a necessity, and households with higher incomes are willing to spend more to improve the quality of water they use. This is consistent with theoretical expectations and justifies the internal validity of the model estimated.
Also, we introduced gender into the model because gender disparity matters in household choices. We found a positive relationship between gender and averting expendi-   Robust standard errors in parentheses: ***p < 0.01, **p < 0.05, *p < 0.1. (Note: coefficients approximate to 4 decimal places.). This evidence is consistent with the a priori expectation.
Again, we introduced a control variable for households with a regular pipe-borne water supply and compared them with households without access to regular pipe-borne water supply, where the latter is used as the reference category. We found a negative and highly statistically significant relation- Again, to inform policy in subsidy allocation to different income groups, the study employs the poverty headcount index by the World Bank for lower middle-income countries and categorizes the respondents into low-income (poor) and high-income (non-poor) groups. Subsequently, we use both parametric and non-parametric tests to investigate whether there is a statistically significant difference between the poor and non-poor averting expenditures.
In Table 7, we find evidence to reject the null hypothesis that there is no statistically significant relationship between low-and high-income groups in their averting expenditures. That is, the high-income group is observed to spend more on averting expenditures than the lowincome group. This study argues that for policy purposes, two scenarios are possible. First, in the case where the poor and the non-poor are exposed to the same sources (e.g., residential pipe-borne system), we expect the poor to consume less relative to the non-poor, hence the government can subsidize the consumption of the poor based on volume used. Second, in the case where both income groups are exposed to different water sources because of the ability to afford (e.g., the high-income group can afford bottled water), the government can subsidize the sources being patronized by the poor.

CONCLUSION
Despite considerable achievement in access to improved water over the years, substantial inequality in the supply of improved water by the GWCL still exists. Using household-level primary data of 1,648 observations, we have identified households' averting behaviours together with associated cost and drivers. This study has shown that, on average, a household is willing to pay Ghs84.30 ($14.70) per month to improve the quality of their domestic water.
Our current estimate constitutes 13.25% of household take-home income.
For policy purposes, this study has provided a demandside estimate necessary for determining the economic viability of connecting households in urban GAR to the regular piped supply network and ensure that improved quality water is served to households. Thus, to permanently end domestic water-borne diseases and provide safe pipe-borne  water at a relatively cheaper price, this study is arguing that households are currently spending an average of 13.25% of their take-home income to enjoy improved water. This estimate depicts households' average WTP for safe pipe-borne water in urban GAR. For the supply of reliable safe pipeborne water, this estimate is very informative for the GWCL and government in determining pricing and subsidy strategies for households, especially the poor and non-poor.
Lastly, the evidence of households' WTP supports the call for the provision of residential reliable safe pipe-borne water for households in urban GAR and by extension, the entire country.

ETHICAL DECLARATION
The data collection process which is part of my PhD research was approved by the Ethics Committee, Faculty of Social Sciences, University of East Anglia, UK. The data were presented at a seminar as well as shared with key supervisors.

DATA AVAILABILITY STATEMENT
Data cannot be made publicly available; readers should contact the corresponding author for details.