The Determinants of Household Water Demand: A Focus on Water, Energy, and Food Prices

1. Introduction

Water is an essential resource for the survival of all living organisms and is crucial in the functioning of almost all economic activities. Increasing concerns about the supply of fresh water and the impact of water shortages on economic activity placed sustainable water management at the top of the global research and policy agenda (Gupta et al. 2020). With households now accounting for a substantial and growing proportion of total water consumption in most developing economies, while household water insecurity persists, municipal water demand has become a principal concern of policymakers in these countries (Heidari et al. 2021; Savelli et al. 2023). As a result, there has been widespread debate regarding municipal water policy design and implementation (OECD 2018; Velasco et al. 2023). A key outcome of this debate has been that a good understanding of the factors that influence household water demand is required to assist policymakers in their decision to implement appropriate policy instruments that lead to sustainable municipal water use and ensure water security (Reynaud And Romano 2018; AbuBakar et al. 2021). This paper therefore analyzes the factors affecting household water demand in South Africa to develop a better understanding of residential water demand in developing country contexts.

Most research analyzing the determinants of household water demand has focused on determining the impact of water price, socio-economic characteristics, housing characteristics, and climatic variables on household water demand (AbuBakar et al. 2021; Nemati and Tran 2022; Yoo et al. 2014; Cominola et al. 2023). The primary focus of this research has been determining the water price elasticity of household water demand (Nauges and Whittington 2010; Abu-Bakar et al. 2021). In terms of price impacts, however, in developing countries, and for poor households, it is not only the impact of water price which affects household water demand. Given the low average incomes of these households, the prices of other basic needs such as food and energy also have a significant effect on household water demand. In South Africa, between 60% and 90% of poor and lower-middle income households’ monthly expenditure is used for water, energy, and food (WEF) purchases (see Figure C.2). The substantial proportion of expenditure used on these goods suggests that when the prices of either of these goods, households will have to alter their consumption of them. A water–energy–food price–consumption nexus thus often exists for poor households. As poor households are particularly vulnerable to WEF insecurity, this nexus can be thought of as a WEF price–security nexus. In this paper, we examine the WEF price–security nexus by determining how changes in the prices of WEF affect household water consumption. This is a crucial and absent aspect of the residential demand literature in developing countries. The results are not only important to developing countries but for all countries because the WEF price–security nexus likely exists for low-income households around the world.

We investigate two primary research questions. The first explores how WEF prices affect household water demand. The second investigates what other factors affect household water demand in the study area. For the second, we focus on the impact of socio-economic characteristics, housing characteristics, and water-saving technologies. While investigating the two research questions, we additionally examine how the estimated effects vary between households using heterogeneous water sources. This is because the analysis of household demand in developing countries is complicated by abundant evidence that, contrary to what is observed in most developed countries, households in these countries have access to and may use more than one of several types of water sources (Zhu et al. 2018; Coulibaly et al. 2014; Nauges and Whittington 2010). Households in our study area possess this unique characteristic, with them having access to ground, rain, and bottled water in addition to tap water. We account for this by running separate regressions for heterogeneous household groups based on the water sources they use. Analysis of household water demand is also complicated by the fact that the determinants of household water demand vary widely between households in different income groups (Zhu et al. 2018). We therefore also run separate regressions for households in heterogeneous income groups. Finally, we use the WEF price and income elasticities estimated to predict how water demand would be affected in various potential scenarios.

2. Literature Review

Since the second half of the 20th century, population growth and climate change, coupled with decreasing freshwater supplies and the ever-increasing cost of water infrastructure, have led to a policy emphasis on water demand management across the globe. This has been particularly important at the household level because households now account for a significant and growing proportion of the total water supply (Heidari et al. 2021; Savelli et al. 2023). In response, an extensive body of literature analyzing the factors affecting household water demand has been developed to support policymakers in implementing sustainable water demand management and promoting water security. Most of the analysis has been done by estimating household water demand functions, in which researchers have focused on determining the impact of water price, socio-economic characteristics (income, age, education, etc.), housing characteristics (size, appliances, gardens, swimming pools, etc.), and climatic variables (rainfall, temperature, etc.) on household water demand (Nauges and Whittington 2010; Yoo et al. 2014; AbuBakar et al. 2021; Oyerinde and Jacobs 2022; Cominola et al. 2023).

In most studies conducted in developed countries, the household water demand function is estimated as a single equation that links tap water consumption (the dependent variable) to water price and a set of explanatory variables that affect demand (Yoo et al. 2014; Reynaud et al. 2018; Suárez-Varela 2020; Reynaud et al. 2018; Suárez-Varela 2020; Oyerinde and Jacobs 2022). This is done to control for preference heterogeneity and other factors which affect household water demand. Utilization of the single equation model implicitly assumes that no substitutes for tap water exist. Unlike in developing countries, water quality and the reliability of water supply services are generally not included as controls in the single equation model because there is little variation between households on the same system (Nauges and Whittington 2010). Instead, the focus is on estimating price and income elasticities of water demand as well as determining the impact of other socio-economic characteristics (Worthington and Hoffmann 2008; Yoo et al. 2014; Reynaud et al. 2018; Suárez-Varela 2020). The primary methodological issues in these studies are price endogeneity and the choice between using marginal or average prices when households face non-linear pricing schemes (Nauges and Whittington 2010). The data used in studies generally come from water utility records. Despite providing panel data, utility records contain little information on the socio-economic and demographic characteristics of households and possess little variation in important explanatory variables such as tariff structure, water quality, and reliability, which hinders the estimation.

The estimation of household water demand in developing countries is complicated by the fact that households in these countries generally have access to, and use more than one type of water source (Nauges and Strand 2007; Coulibaly et al. 2014). These water sources include in-house tap connections, wells, boreholes, water vendors, natural water sources, and rainwater, amongst others. This creates a problem for the single equation model which implicitly assumes that no substitutes for tap water exist. Other problems encountered when conducting household water demand studies in developing countries include the fact that households with private tap connections generally have unmetered connections and where there are meters, they are often unreliable (Nauges and Whittington 2010; Zhu et al. 2018). In addition, data sets which contain information on the amount of water used are often unavailable. Researchers therefore generally use cross-sectional data from household surveys which limits the identification of causal relationships (Nauges and Whittington 2010).

The predominant focus in both the developed and developing country literature has been on determining the water price elasticity of household water demand (Worthington and Hoffmann 2008; Zhu et al. 2018; Abu-Bakar et al. 2021). This is because price represents one of the most predominant tools used to manage the demand for water (Arbués et al. 2003; Ghimire et al. 2016; Ščasný and Smutná 2021). The essential logic flows from the law of demand, which argues that the demand for water should be inversely related to water price, which would make sense if water were treated as a pure economic good. However, as Savenije and van der Zaag (2002) noted, the relationship between water consumption and price is starkly different from that of a normal economic good because water is often irreplaceable.

A synoptic survey of residential water demand studies conducted between 1980 and 2008 in developed countries revealed that the water price elasticity was inelastic and varied between −0.25 and −0.75 (Worthington and Hoffmann 2008). Most studies in developed countries have continued to produce similar findings (Yoo et al. 2014; Reynaud et al. 2018; Suárez-Varela 2020), although higher price elasticities have been reported with increased frequency. This has been directly attributed to increased outdoor water use, with price elasticity estimates now ranging from −1.57 to −3.33 when outdoor uses are considered (Klaiber et al. 2014; Yoo et al. 2014; Kotagama et al. 2017). Most of the price elasticity estimates for households in developing countries, with a private water connection, were also inelastic and ranged from −0.3 to −0.6 (Nauges and Whittington 2010). This is similar to more recent research which has estimated price elasticities from a piped connection of between −0.2 and −0.7 in developing countries (see Table C.4.). When accounting for multiple water sources, the price elasticity estimates have sometimes been shown to be substantially larger, ranging from −0.62 to −2.33 (Coulibaly et al. 2014).

In terms of the price impacts on water demand in developing countries, and for poor households, we theorize that it is not only the impact of water price which affects household water demand. Given the high proportion of expenditure used for WEF purchases by these households (see Section 3), we believe that price changes in either of these goods will impact household water consumption and security. This has been an overlooked aspect in research analyzing household water demand and should be accounted for going forward. We close this gap in the literature and include WEF prices within our estimation procedure and examine their effects on household water consumption in terms of the WEF price–security nexus (see Section 3).

3. The Water–Energy–Food Price–Security Nexus: Theoretical Framework

Approximately one-third (30%) of household expenditure of countries in the European Union (EU) is used for WEF purchases (Eurostat 2024). The theory would suggest that, in developing countries and for poor households, the proportion of expenditure attributed to these goods would be significantly higher given their lower incomes. Focusing on food expenditure alone, households in Africa have been shown to use between up to 60% of their total expenditure for food purchases (USDA Economic Research Service (ERS) 2023). For example, the proportion in Nigeria is 59% and in Kenya it is 57% (USDA Economic Research Service (ERS) 2023). By comparison, the proportion of expenditure on food is only around 20% in South Africa on average. This, however, is driven by rampant inequality with the top 10% of income-earning households in South Africa heavily swaying the average. Poor South African households (bottom 10%) have been shown to spend close to 90% of their income on WEF purchases (see Figure C.2). Similarly, lower-middle-income households spend about 60% of their income on these resources.

Given the large proportion of expenditure used for WEF purchases by poor and lower-middle-income households (see Figure C.2), any change in their prices would likely have a significant effect on the real household income available. To compensate for increases in the prices of any of these resources, in addition to altering the consumption of the good that had a price increase, households will likely have to alter their consumption of the other resources. For example, food price increases could impact water or energy consumption (Ningi et al. 2021). This is because poor households generally consume only enough food to survive (Brewis et al. 2020). When food prices increase, they thus have limited scope to reduce their consumption further or change to less expensive foods. Given this, they may have to instead decrease consumption of water and/or energy. Similarly, energy prices could impact water and/or food consumption. Given the low levels of WEF consumption in poor households (Ningi et al. 2021), these price changes impact household resource security. We thus theorize that a water–energy–food price–security nexus exists for poor households (see diagram in Figure 1).

In the context of household resource consumption and security, previous research has not accounted for the water–energy–food price–security nexus. For example, in addition to the inclusion of control variables, household food consumption and security studies have focused on the impact of food prices (Andreyeva et al. 2010; TaghizadehHesary et al. 2019; Huang et al. 2022). Similarly, water studies focused on water price impacts (Zhu et al. 2018; Abu-Bakar et al. 2021; Ščasný and Smutná 2021), and energy studies on energy price impacts (Taghizadeh-Hesary et al. 2019; Aydin and Brounen 2019; Chindarkar and Goyal 2019). We suggest that studies aiming to estimate water, energy, or food consumption or security be extended to include the prices of all three goods simultaneously. In this paper, we do this and analyze the WEF price–security nexus in terms of water consumption, by determining the impact of WEF prices on household water demand (see diagram in Figure 2).

4. Data

The primary data used in this study was obtained from a household survey conducted in early 2019 in the Mbombela Municipality of South Africa, which is in the Mpumalanga Province (see map in Figure C.4). To ensure a representative sample was obtained a stratified sampling technique was used during the data collection process. Participating households were asked a set of approximately 40 questions which allowed us to obtain comprehensive data on important variables for estimating the determinants of household water demand (the questionnaire is provided in Appendix A). Following data collection, we undertook a data cleaning process to prevent biases and inconsistencies caused by data capture errors (measurement errors and outliers (see Appendix D). The final sample used in the analysis consisted of 472 households.

We supplemented the household survey data with food and energy price data from Statistics South Africa (Stats SA) to estimate the food and energy price elasticities of household water demand. We obtained this data for the months of December 2018, and January and February 2019 as these were the months for which we collected tap water bill and consumption data. The price data we use corresponds to the monthly consumer price indexes (CPI) for food and energy in South Africa, respectively. For energy, we include electricity and other fuels used. The key dependent variable (tap water price) was determined by following Deaton (1990) and estimating its unit values. To do this we divided a household’s monthly water bill by their respective monthly tap water consumption (the independent variable). Household income was calculated by summing together all household income sources (see income sources in the questionnaire Appendix A).

We supplemented the household survey data with food and energy price data from Statistics South Africa (Stats SA) to estimate the food and energy price elasticities of household water demand. We obtained this data for the months of December 2018, and January and February 2019 as these were the months for which we collected tap water bill and consumption data. The price data we use corresponds to the monthly consumer price indexes (CPI) for food and energy in South Africa, respectively. For energy, we include electricity and other fuels used. The key dependent variable (tap water price) was determined by following Deaton (1990) and estimating its unit values. To do this we divided a household’s monthly water bill by their respective monthly tap water consumption (the independent variable). Household income was calculated by summing together all household income sources (see income sources in the questionnaire Appendix A). We convert the WEF price values as well the income values into log form in the regression specifications to directly determine their respective elasticities. An important consideration when introducing certain explanatory variables into one’s model is that opinions about water quality and water scarcity should be used with caution as they could introduce endogeneity into the model (Nauges and Whittington 2010; Zhu et al. 2018). For example, households that have experienced diseases caused by bad water quality may be more likely than others to believe the water is of poor quality and hence may show different water use behavior (Nauges and van den Berg 2009). In addition, beliefs about water quality could be correlated with education and income which causes collinearity issues (Whitehead 2006). To avoid these problems, we follow Nauges and Whittington (2010) and develop average opinions on water quality and water scarcity for households living in the same area to prevent potential biases caused by individual households.

The average tap water consumption of households in our study area was 2,268 L per month (lpm) (see Table 1). Consumption was at similarly low levels for all household income groups, although rich households consumed 20% more tap water than poor households. The average tap water consumption for households in our study equates to 19 L per capita per day (lpcd). This amount is just below the lower end of the 20–40 lpcd basic water requirement target set by the USAID, World Bank, and World Health Organization, which indicates that households were likely to only meet their basic water needs through tap water consumption. The low tap water consumption levels are due to the Mpumalanga province being a highly water-scarce region, which experienced low rainfall and had low dam levels during the study period (Ebhuoma et al. 2020; Department of Water and Sanitation (DWS) 2023). The conditions would worsen and result in a severe drought in Mpumalanga a few months later (Association for Water and Rural Development (AWARD) 2019; Department of Water and Sanitation (DWS) 2023). As our study was conducted in a highly water-scarce area and during a time of water shortages, one should note that households would have had substantially lower tap water consumption levels than seen in other areas. There are two other things to note about the low tap water consumption levels. First, they will impact the magnitude of the effects found because any change in tap water consumption would have a substantially larger proportionate effect on total demand than in studies with higher household water consumption levels. Second, most (93%) households in our study supplemented their tap water use with the use of other water sources (rainwater, groundwater, and bottled water). The use of these additional water sources would have allowed households to have higher overall water consumption than the tap water consumption levels reported (descriptive statistics by water sources used are presented in Table C.5).

The household-level data obtained from the survey are of benefit to this study for several reasons. First, household-level data are highly preferable for estimating the determinants of household water demand (Nauges and Whittington 2010; Abdullah et al. 2019). Second, we collected three months of tap water bill and consumption data which allowed us to overcome issues faced when using cross-sectional data from household surveys (as explained in Section 2). Third, unlike in most developing country studies, we obtained accurate water consumption data as households in our study area had tap water meters. Fourth, we collected comprehensive data on key socio-economic and housing characteristics of households that affect water demand. This is important because accurate data that combines water consumption, billing, socio-economic characteristics, and housing characteristics is often very limited. Fifth, we collected data on the water-saving strategies and technologies used by households as well as their attitudes toward water conservation practices. This information is important as these are factors that are key determinants of household water demand (Syme et al. 2004; Hoolohan and Browne 2016).

5. Method

5.1. Estimation strategy

We model household water demand as a behavioral phenomenon which is modeled as a final good for consumption (Zhu et al. 2018). Researchers using this approach have generally relied on regression techniques to estimate household water demand (David and Inocencio 1998; Kotagama et al. 2017; Oyerinde and Jacobs 2022). In regression analysis, an equation relating household water consumption, the independent variable, to multiple demand-related variables is estimated to determine their impact on household water consumption. This process allows one to isolate the effect of, for example, a change in price on the quantity of water demanded, while controlling for other variables that factor into the consumer’s decision process. Most previous studies have used conventional regression methods such as ordinary least squares (OLS), two- and three-stage least squares (2SLS and 3SLS), and time series analysis to estimate the determinants of household water demand (Billings and Agthe 1980; Ghimire et al. 2016; Oyerinde and Jacobs 2022). We follow this literature and use regression techniques to estimate household water demand in our study area.

Before running regressions, we are faced with two problems due to households in our study area experiencing a non-linear pricing scheme. First is the choice between using marginal or average water prices in our regression specifications. Many researchers use average price despite the use of marginal price being suggested by economic theory (Ghimire et al. 2016; Pérez-Urdiales and García-Valiñas 2016). This is because households are seldom aware of the tariff structure they face and hence are more likely to respond to changes in average price than marginal price (Nauges and Whittington 2010). Households in our study indicated that they were not well informed about the pricing structure and hence we use average price. Second, is the issue of endogeneity in the tap water price variable. Endogeneity is a problem because it results in biased and inconsistent estimates which limits the interpretation of one’s results (Hill et al. 2021). Endogeneity in the case of the nonlinear pricing scheme is caused by simultaneity because just as tap water price is a determinant of water consumption (the independent variable), the level of water consumption determines the price of tap water. Endogeneity in the tap water price has also often been linked to measurement error, which has plagued the use of unit values in determining price elasticities from household surveys (Deaton 1990; Nauges and van den Berg 2009). The unit values for water price in this paper, however, were calculated using consumption data from accurate water meters and data from household water bills which limits the possibility of it causing measurement error (Nauges and van den Berg 2009). Endogeneity issues in this type of analysis are also linked to potential omitted variable bias (Hill et al. 2021). To address the endogeneity issue caused by the non-linear pricing structure and potential omitted variable bias in the specification, we run a 2SLS regression procedure with the use of instrumental variables (IVs) (Ghimire et al. 2016; Ščasný and Smutná 2021). As is often done (Zhu et al. 2018; Abdullah et al. 2019), we run an OLS regression as a first step to both test for endogeneity and allow for a comparison of the OLS and 2SLS results.

As highlighted earlier, and often seen in developing countries, households in our study area had access to and used more than one type of water source. In addition to tap water, other water sources used included ground water (boreholes and wells), rainwater, and bottled water. Unfortunately, no accurate data on water consumption or the price of non-tap water sources was available and thus we cannot determine the price elasticities of demand for these different water sources. We do, however, account for household’s heterogeneous use of water sources, which is something of particular importance in developing countries (Nauges and van den Berg 2009; Coulibaly et al. 2014). To do this, we estimate separate equations for five heterogeneous household groups determined by the water sources they use. These household groups are — (1) tap water only, (2) tap and others, (3) tap and ground, (4) tap and rain, and (5) tap and bottled. This allows us to determine the extent to which having access to and using other water sources affects the determinants of household water demand. In addition to estimating separate equations for households using different water sources, we estimate the demand equation for households’ heterogeneous income groups. We develop four income groups based on the income quartiles found in the data set. The groups we use are — (1) poor (income < R 14,200), (2) lower-middle (R 14,200 ≤ income < R 22,400), (3) upper-middle (R 22,400 ≤ income ≤ R 35,000), and (4) rich (income > R 35,000). This allows us to determine how the factors affecting household water demand vary between households in different income groups. Such analysis is important in both developed and developing countries, particularly in ones with high inequality such as South Africa, because impacts can vary widely by income group (Zhu et al. 2018).

5.2. Econometric specification

Conventionally, the water demand function is specified as $Q=f(P$,$X)$, where Q represents the household water consumption level, P symbolizes the price of water, and X represents a vector of other exogenous variables such as household income, household size, education level, etc. (Coulibaly et al. 2014; Oyerinde and Jacobs 2022). As a first step, we run an OLS regression to estimate the determinants of household water demand. We initially include a large set of variables suggested by the literature, and which are available in our data set, in our model as well as including food and energy prices. After testing the significance of the model and the included variables, the final OLS regression we use is specified in the following equation:

We suspect that the tap water price variable in our study is endogenous and hence correlated with the error term ${\epsilon }_i$. To prevent biased and inconsistent estimates, we then employ a 2SLS estimation procedure with the use of IVs (Kotagama et al. 2017; Oyerinde and Jacobs 2022). Identifying suitable IVs is the key challenge of the 2SLS model because a good instrument (Z) must satisfy three important conditions. First, the IV (Z) must have a causal effect on the endogenous variable (X), the inclusion restriction. Second, Z may affect the outcome variable (Y) only through its effect on X. This ensures that Z does not have a direct influence on Y, which is referred to as the exclusion restriction. Third, there can be no confounding for the effect of Z on Y (Martens et al. 2006; Wunsch et al. 2006). A wide variety of IVs have been used in previous water demand studies, such as annual household expenditure on water, the marginal water price in each consumption block, the difference between each successive block price, fixed charges, locality, and water scarcity, amongst others (Nieswiadomy and Molina 1989; Reynaud et al. 2005; Nauges and van den Berg 2009; Price et al. 2014; Pérez-Urdiales and García-Valiñas 2016; Clarke et al. 2017). After much research and testing, we identified three appropriate IVs. For the first two, we follow Wichman et al. (2016) and many others and use fixed service fees and marginal prices (i.e., price for each consumption block) as these instruments are uncorrelated with water consumption but highly correlated with average price (Nieswiadomy and Molina 1989; Reynaud et al. 2005; Olmstead 2009; Price et al. 2014; Pérez-Urdiales and García-Valiñas 2016). For our third IV, we follow Nauges and van den Berg (2009) and choose household location dummies as identifying instruments for the water price paid by the household. We have nine such location dummies. These dummies are good instruments as long as household’s location choice does not depend on the price of water. We believe this to be a reasonable assumption for this study area as it is highly unlikely that household’s decision to live somewhere would be based on differences in water price. A detailed description of the IVs chosen, and the testing procedure followed is provided in Appendix B.

As the name suggests, the 2SLS regression procedure is conducted in two stages. First, the endogenous variable (tap water price) is regressed against the chosen IVs. The tap water price predicted in the first stage is then regressed against a set of predictor variables and the instruments in the second stage. The instruments are used to create an accurate proxy for the endogenous variable, which provides consistency in the limits estimated. In the first stage we regress tap water price (the endogenous variable) against the chosen IVs (Fixed Fees, Marginal Prices in Each Block, Locality). To do this, we estimate the following equation :

In stage two of the 2SLS estimation procedure, the new tap water price $P^{\ast }$ (estimated in stage one) is then regressed against all the exogenous variables and the IVs. For this, we estimate the following equation:

where

To ensure the 2SLS model is the correct model to use, we test for endogeneity. To do this, we run the Durbin–Wu–Hausman test which detects endogenous regressors in a regression model. The results of this test showed that there was endogeneity and hence the 2SLS regression model is the correct model to use. Another common problem when estimating household water demand is multicollinearity caused by the correlation between the explanatory variables (Dandy et al. 1997; Oyerinde and Jacobs 2022). For example, income may be correlated with education level or owning a swimming pool. This creates a problem for the least squares estimates as it leads to large standard errors which affects the significance of the model, its included variables and means. As a result, one cannot estimate household water demand precisely. To ensure there are no multicollinearity issues present in our specifications, we check the variance inflation factor (VIF). The VIF values we find are between 1.04 and 1.42. Most researchers suggest that VIF values larger than 5 indicate severe collinearity and we are thus confident that multicollinearity is not an issue in our specification (Basu et al. 2017; Oyerinde and Jacobs 2022). For sensitivity analysis, we do two things. First, in addition to running an OLS and 2SLS model, we estimate a fixed effect (FE) model. This is done to test the robustness of the results by analyzing if the trends seen in the OLS and 2SLS models hold in the FEs model. Second, to ensure the water price elasticity estimates are robust we also vary the tap water price variable. For this, we change the tap water price variable from being the average price paid per liter of tap water (which includes the fixed fee) to the average price paid per liter without the fixed fee.

5.3. Scenarios

Once the determinants of household water demand have been estimated, we investigate how the estimated elasticities would affect household water demand in various scenarios. To do this, we create four scenarios in which we alter WEF prices as well as household income to see how household water demand would be affected in each. We create two scenarios for the time of study (2019), and two potential scenarios for the year 2035. A description of each scenario and is provided in Table 2, while the methodologies used for their calculations are provided in Table C.12.

6. Results

6.1. WEF price elasticities

6.1.1. Overall and by income group

The average water price elasticity for households in our study area was negative and statistically significant at the 1% level (see Table 3). This shows that households reduced their tap water consumption in response to price increases. The magnitude of the elasticity was −0.672 which implies that a 10% increase in the tap water price would lead to a 6.72% decrease in household tap water consumption, ceteris paribus. This value is in line with the literature which generally estimates a water price elasticity of −0.2 to −0.7 in developing countries (see Table C.4.). The tap water price elasticity was negative across all household income groups, although it was only statistically significant for poor households and those in the upper-middle income group. The elasticity for poor households was the highest at −0.935, which suggests that a 10% increase in tap water price would result in a 9.35% decrease in water consumption, ceteris paribus. This shows that poor households were significantly more sensitive to changes in tap water prices than other households, a trend often seen in the literature (Mayol 2017; Masayoshi and Sunaga 2021). For rich and lower-middle-income households, the insignificant results show that changes in tap water price did not influence their water consumption (see Section 6 for a discussion on this).

The food price elasticity was −0.174, which suggests that a 10% increase in the CPI of food would decrease tap water consumption by 1.74% (see Table 3). The food price elasticity was negative and statistically significant for all household income groups, except rich households, which shows that food price increases led to decreased tap water consumption for these households. For the non-rich households, a 10% increase in the CPI of food was shown to decrease tap water consumption by between 2.11% and 3.07%. For energy, the price elasticity was highly inelastic at −0.072, which implies that a 10% increase in the CPI of energy would only decrease tap water consumption by 0.72%, ceteris paribus. As was seen for the water price elasticity, the energy price elasticity was only significant for poor households and those in the upper-middle income group. The magnitude of the energy price elasticity was particularly significant for poor households (−0.163), with the results showing that a 10% increase in the CPI of energy led to a decrease in household water consumption of 1.63%.

6.1.2. By water sources used

The water price elasticity was negative for all household groups irrespective of the water source(s) they used, although it was only statistically significant for households which used tap and other water sources, except for those which used tap and groundwater (see Table 4). For household groups with statistically significant water price elasticities, this result implies that tap water consumption decreased when its price increased. The magnitude of the effect was largest for households which used tap and bottled water (−0.794), in comparison to those which collectively used tap and other water sources (−0.597) and those which used only tap and rainwater (−0.543). The elasticities found were in line with the estimates from the literature for developing countries (see Table C.4.). The food and energy price elasticities were not statistically significant for households which used tap water only (see Table 4). For households which supplemented their tap water consumption with the use of other sources, the food and energy price elasticities were negative and statistically significant at the 1% level. This suggests that an increase in the CPI of food or energy would lead to a decrease in household tap water consumption for these households. The magnitude of the food price elasticity (−0.199) for these households was approximately three time larger than for energy (−0.0737), which indicates that these households were significantly more sensitive to food price changes than energy price changes. These findings are in line with what was shown in Section 5.1.1.

6.2. Other determinants of household water demand

6.2.1. Socio-economic characteristics

The overall income elasticity of water demand for households in our study was positive and statistically significant at the 1% level (see Table 5). This indicates that tap water consumption increased as household income increased. The magnitude of elasticity was 0.194 which implies that a 10% increase in household income increased tap water consumption by 1.94%, ceteris paribus. This value is in line with the literature which generally estimates an income elasticity of 0.1–0.4 for households in developing countries (see Table C.4.). The income elasticity was only statistically significant for poor (0.153) and rich households (0.474). Therefore, only these households were shown to increase tap water consumption when their income increased. The magnitude of the elasticity for rich households was approximately three times larger than for poor households. The income elasticity of water demand for households which used tap and other water sources was positive and statistically significant (see Table C.3.). For these households, increases in income resulted in increased tap water consumption. By sub-group, the results were only significant for households which used tap and rainwater (0.203) and those which used tap and bottled water (0.270). The income elasticities for these households were in line with estimates from the developing country literature of between 0.1 and 0.4 (see Table C.4.).

Overall, the coefficient for household size was positive and statistically significant for households in our study (see Table 5). This implies that having an additional household member increased water consumption which is to be expected (Cheesman et al. 2008; Garcia et al. 2019). The effect, however, was only marginal at 0.0456, which suggests that having an additional household member increased tap water consumption by 4.56%. Results by household sub-group revealed that the effect was only statistically significant for households which used tap water only (see Table C.3.). The magnitude of the effect for these households was 0.24, implying an additional family member increased household water consumption by 24%, ceteris paribus. Overall, the education level of the household head did not have a significant effect on tap water consumption for households in our study area (see Table 5). These findings are in line with evidence from the literature which generally does not show education to be a significant predictor of water demand, with it instead arguing that education level is to be related to environmental consciousness and awareness, as opposed to directly influencing water consumption, or to the decision to improve diversification in water supply sources (Syme et al. 2004; Larson et al. 2006; Dieu-Hang et al. 2017). There was, however, a statistically significant effect seen for households which used tap and rainwater (−0.0179) and those which use tap and groundwater (0.0428).

6.2.2. Housing characteristics and water saving technologies

Overall, owning a swimming pool was shown to increase household tap water consumption by 22.2% (see Table 6). This positive effect is to be expected given the large water requirements of swimming pools and is a finding congruent with the literature (Balling et al. 2008; Morote and Hernández 2016). Like swimming pool ownership, the presence of a lawn, trees, or garden was shown to positively influence household water consumption (see Table 6), a common finding in the literature (Runfola et al. 2013; Hussien et al. 2016). Specifically, it was shown to lead to an increase in tap water consumption of 13.9%. As often seen in the literature (Willis et al. 2013; Garcia et al. 2019; Leonard and Gato-Trinidad 2021), the use of water saving technologies (rainwater tanks, rainwater systems, and efficient washing machines) was shown to have a significant negative effect on household tap water consumption (see Table 6). The use of rainwater systems and rainwater tanks were shown to decrease household tap water consumption by 32.9% and 21.6%, respectively, ceteris paribus. This is a trend often found in the literature (Willis et al. 2013; Umapathi et al. 2013; Leonard and Gato-Trinidad 2021), although the magnitude of the effect was larger in our study. The impact of using rainwater technologies was significant for all household income groups, except poor households. This is likely because these households could generally not afford these technologies. Overall, the use of efficient washing machines was shown to decrease household tap water consumption by 23.1%. This is a substantial effect, which highlights the water savings these appliances provide. Willis et al. (2013) presented similar findings by showing that efficient washing machines used four times less water than standard ones and could lead to monthly decreases in water consumption of around 1,000 L per household.

6.3. Sensitivity analysis

For sensitivity analysis, in addition to the OLS and 2SLS regressions, we ran a FEs model. The results from this analysis revealed that the trends found in the OLS and 2SLS regression models held in the FEs models (see Tables C.8–C.10). First, this was seen for the water, food, and energy price elasticities which were negative and statistically significant at the 1% level, as well as being of a similar magnitude to the 2SLS model. This implies that an increase in any of these prices would lead to a decrease in household tap water consumption. Second, the income elasticity was positive and statistically significant in all three models, which confirms that increases in income resulted in increased household tap water consumption. Third, the result trends (sign, significance, and magnitude of coefficients) for the other determinants of household tap water consumption found in the FE model were in line with the findings of the OLS and 2SLS model (see Tables C.8–C.10). To ensure the water price elasticity estimates were robust we also varied the tap water price variable, as highlighted above in (see Section 4.2). The results of this test produced findings in line with the original 2SLS estimation procedure (see Table C.11). These findings highlight the robustness of our results.

6.4. Scenarios

The basic water requirement for water security set by the USAID, World Bank, and World Health Organization is 20–40 lpcd. The only scenario in which households achieve this level of water security through tap water consumption alone is the FREEW scenario (33 lpcd). This shows that the provision of free water and electricity to acceptable thresholds would promote household water security, particularly for poor households (see Table 7). In this scenario, average household tap water consumption was shown to increase by 74% from its study period level of 2,262–3,955 lphm, while for poor households it was shown to increase by 110% to 4,406 lphm. Poor households were the least likely to own rainwater tanks or have groundwater access (see Table 7) and thus it is particularly important for them to achieve the water security level through tap water consumption. In all scenarios, except FREEW, household tap water consumption was shown to be negatively affected by the price and/or income changes implemented in the scenarios. Average tap water consumption decreased the most in the CC 2035 scenario (−57%), compared to a 33% and 9% decrease in the BAU 2035 and FOOM scenarios, respectively. Poor households tap water consumption decreased more significantly than others in these scenarios, which shows their sensitivity to price changes. Tap water consumption decreased to extremely low levels in the CC 2035 scenario with consumption equal to 8.1 lpcd on average and only 4.5 lpcd for poor households. These levels indicate severe water insecurity. High water scarcity was also seen in BAU 2035 scenario, which indicates that policy measures will need to be implemented in the future to ensure water security in the study area.

7. Discussion

Our findings show that increases in water, food, or energy prices led to a decrease in tap water consumption (see Table 3). This highlights the substitution effects between tap water consumption and the consumption patterns for food or energy when their respective prices change. Previous studies have primarily focused on estimating the water price elasticity of household water demand, however, we show that in developing countries, and for poor households elsewhere, the prices of other essential goods such as food and energy also have an important impact on household water consumption.

The magnitude of the WEF price effects was largest for poor households which highlights their sensitivity to price changes. Rich household’s water consumption was not impacted by changes in water, food, or energy prices. We argue that this was because their high-income levels imply that the price changes in these goods were not large enough to significantly influence their tap water consumption levels. The water and energy price elasticities were insignificant for lower-middle-income households (see Table 3). The reason, we believe this occurred was because these households were already consuming extremely low levels of tap water (see Table 1) and hence were unable to reduce their consumption further in the face of price increases. It is more likely that water and energy price increases led them to alter their consumption of food or energy.

Given the high tap water prices in our study area (see Appendix E), one would expect a significantly larger tap water price elasticity than is commonly seen due to the larger real price impacts. In addition, the availability of additional water sources would potentially lead to higher tap water price elasticities (Coulibaly et al. 2014). The elasticities we find are at the higher end of the range of previous estimates (see Table C.4.), however, they are not excessively high. We believe there are two reasons for this. First, given the water scarcity in the study area, household’s sensitivity to prices may have decreased, something which has been shown in previous research (Garrone et al. 2019). Second, by using three months of data, we focused on estimating short-run price elasticity which is generally lower than long-term elasticity estimates because fixed variables in the short run become more elastic in the long run (Lim et al. 2012; Buschsbahm 2022).

The tap WEF price elasticities were significant for households that used tap and other water sources and insignificant for those that used tap water only (see Table 3). We argue this is because households that had access to other water sources could use more water from these sources when prices increased, while households using only tap water could likely not reduce their consumption (see Section 3). This is because their low consumption levels suggest they were using only enough water to meet basic requirements and hence could not reduce consumption further in the face of WEF price increases.

The income elasticity of water demand was only statistically significant for poor and rich households (see Table 4). This may be due to rich households acquiring water-intensive appliances and/or other water-intensive things such as gardens and swimming pools when their income increased. For poor households, it was likely because they could only afford enough water to meet their basic needs (see Table 1). When their income increased, they could then increase water consumption to levels beyond this. The reason for insignificant income elasticities for households in the lower-and-middle income groups may be due to increases in income being used to increase consumption of alternate water sources and hence not affect their tap water consumption.

The use of water saving technologies was shown to lead to large decreases in household water consumption (see Table 7), with the magnitude of the effect found substantially larger than in other studies (Willis et al. 2013; Umapathi et al. 2013; Leonard and Gato-Trinidad 2021). This was due to the low tap water consumption levels of households in our study, which implied that any use of water saving technologies allowed for a more substantial proportionate effect on household tap water consumption.

8. Conclusion

As evidenced by SDG 6, sustainable water management has been placed at the top of the global research and policy agenda. A primary concern of policymakers is managing municipal water demand as it accounts for a substantial and growing proportion of total water consumption in most developing economies, yet household water insecurity persists (Gupta et al. 2020; Heidari et al. 2021; Savelli et al. 2023). Given this, a good understanding of the factors that influence household water demand is required to assist policymakers in their strategy of implementing appropriate policy instruments that lead to sustainable municipal water use and ensure water security. In this paper, we analyzed the factors affecting household water demand in the Mpumalanga province of South Africa to develop a better understanding of the determinants of residential water demand in developing countries. We investigated two primary research questions. First, we explored how WEF prices affect household water demand. Second, we explored what other factors affected household water demand in the study area, focusing on income, other socio-economic characteristics, housing characteristics, and water saving technologies. Further, we explored how the effects changed between heterogeneous household groups based on their (1) income levels, and (2) the water sources they used.

The results revealed that increases in water, food, or energy prices led to a decrease in tap water consumption. This highlights the substitution effects between tap water consumption and the consumption of food or energy when their respective prices change. Our findings support the theory of the WEF price–security nexus, particularly in the context of poor households (see Section 3). This is a crucial finding for policymakers because it highlights the need for consideration of the impact of energy and food prices on household water demand. Previously the focus has been on the impact of water price alone, however, we show this needs to be extended and suggest policymakers account for the WEF price–security nexus going forward. Results also revealed that having access to other water sources allowed households to reduce their tap water consumption in the face of price increases, while households without alternate water sources were unable to do so. Finally. It was shown that the use of water saving technologies had a significant negative effect on household tap water consumption. Policymakers should design water management strategies with this in mind and promote/support the adoption of these technologies.

We acknowledge a few limitations of this study. First, having only three months of data is a limitation of this study (see Section 7). By having a longer time frame, one would be able to determine the longer-term consumption responses to price changes. We suggest future research use a longer time frame where possible. Second, we note that the water prices used in this paper were impacted by corrupt government practices which likely effected the water price elasticities calculated (see Appendix E). Third, this study was conducted in a drought period which meant that tap water consumption levels were extremely low. This would have impacted the magnitude of the effects found in this study. Going forward, we recommend that more research is done in developing countries to determine how the prices of WEF relate to household WEF consumption and security.

Appendix A. Survey Questionnaire Used

QUESTIONNAIRE

Introduction

Good morning/afternoon/evening. My name is XXXX, and I am part of a research team collecting data in Cape Town and Mpumalanga. The data collection is part of collaboration between the Sultan Qaboos University in Oman and the University of Cape Town in South Africa. The project’s main objective is to understand the challenges faced by water users. From this city, a small sample of households has been selected and I am now in the process of discussing with people like you to get information about water management behavior in your community. This information is confidential and will only be used for the purposes of this study, which will not make reference by name to any one respondent. I will be grateful if you could assist me in filling out this questionnaire in as honest as possible. The interview will take about 40 min of your time and participation is voluntary. Your responses will help the City of Cape Town to design better water policy measures.

Section A: Identification		Code
A1	Questionnaire number
A2	Date
A3	Suburb
A4	Enumerator
A5	GPS coordinates

Section B: General information on household water consumption			Code
B1	Over the past 3 months, what was the household’s tap water consumption?	Current month
		Previous month 1
		Previous month 2
B2	Over the past 3 months, what was the household’s monthly tap water bill?	Current month
		Previous month 1
		Previous month 2
B3	Do you know how water is being priced in your city? 0=no 1=yes
B4	Does household drink tap water? 0=no 1=yes
B5	(a) In your opinion, how critical do you think the water situation in the city is? Please rate 1 to 10
	(b) Number of times household experienced water interruption past 12 months
B6	Please rate your tap water quality using a scale from 1 to 10
B7	Does household buy bottled water for drinking purposes? 0=no 1=yes
B8	(a) Does household own a borehole? 0=no 1=yes
	(b) How much water can be taken from the borehole per day in litters?
	(c) Does household own a well point? 0=no 1=yes
	(d) How much water can be taken from the well point per day in litters?
B9	Does household use tap water for irrigation on the property?0=no 1=yes
B10	Does household own a lawn? 0=no 1=yes
B11	Does household own trees?0=no 1=yes
B12	Does household own a flower garden?0=no 1=yes
B13	Does household own a vegetable garden? 0=no 1=yes

Section C: Socio-Economic Variables		Code
C1	Age of household head
C2	Gender of household head 0=Female 1=Male 2=Not disclosed
C3	(a) Education level of household head 0=None 1 = Primary 2=Secondary 3=Tertiary
	(b) Number of years in school for the household head
	(c) Education level of spouse 0=None 1 = Primary 2=Secondary 3=Tertiary
	(d) Number of years in school for the spouse
C4	(a) Is household head employed?0=no 1=yes
	(b) Is spouse employed? 0=no 1=yes
	(c) Number of employed household members
C6	(a) Employment income of household head (Rand)
	(b) Employment income of other household members (Rand)
	(c) Income from business (Rand)
	(d) Remittances (Rand)
	(e) Pension
	(f) Government transfers
	(g) Income from other sources (Rand)
C7	Household size
C8	Number of children under five
C9	Does household own a car? 0=no 1=yes
	Number of cars owned
C10	Does household own a swimming pool?0=no 1=yes
	How big is the swimming pool in litters

Section D: Household water saving strategies and technologies			Code
D1	Is household aware of any water saving strategies?0=no 1=yes
D2	(a) Use shower rather than the bath tab0=no 1=yes
	(b) Turning off the tap when soaping up in the shower 0=no 1=yes
	(c) Turning off the tap when washing dishes0=no 1=yes
	(d) Reducing the number of baths/shower0=no 1=yes
	(e) Reducing the length of baths/shower0=no 1=yes
	(f) Reducing toilet flushes0=no 1=yes
	(g) Use bucket for bathing0=no 1=yes
	(h) Turn off the tap when cleaning teeth0=no 1=yes
	(i) Use bucket for watering the garden rather than horse pipe 0=no 1=yes
	(j) Use bucket for washing car(s) rather than horse pipe0=no 1=yes
D3	Is household aware of any water saving technologies?0=no 1=yes
D4	Has household installed water saving technologies?0=no 1=yes
	Type of water saving technology	1. Efficient shower head 0=no 1=yes
		2. Toilet flushing system0=no 1=yes
		3. Efficient washing machine 0=no 1=yes
D5	Does household use graywater technologies? 0=no 1=yes
D6	Type of graywater technology	1. Bucket 0=no 1=yes
		2. Simple graywater tank 0=no 1=yes
		3. Graywater system installed 0=no 1=yes
		4. Pond 0=no 1=yes
D7	Does household use rainwater technologies? 0=no 1=yes
D8	Type of rainwater technology	1. Bucket0=no 1=yes
		2. Simple rainwater tank0=no 1=yes
		3. Rainwater system installed 0=no 1=yes
		4. Pond0=no 1=yes

Section E: Household attitudes towards water conservation			Code
E1	Do you think households in your area should prioritize water conservation? 0=no 1=yes
E2	Should household be allowed to harvest and use rainwater?0=no 1=yes
E3	Should household be allowed to harvest and use waste water?0=no 1=yes
E4	Do you think rainwater is associated with environmental/health hazards? 0=no 1=yes
E5	Do you think graywater is associated with environmental/health hazards? 0=no 1=yes
E6	Do you think the use of rain helps to save tap water?0=no 1=yes
E7	Do you think the use of graywater helps to save tap water?0=no 1=yes
E8	Do you think rain can help to reduce your water bill?0=no 1=yes
E9	Do you think graywater can help to reduce your water bill?0=no 1=yes
E10	Do you think tap water is correctly priced by city authorities?	1. Too low 0=no 1=yes
		2. Correctly priced 0=no 1=yes
		3. Too high 0=no 1=yes
E11	How much are you willing to pay to enjoy improved water services and quality in your city?

Section F: Constrains to adoption of water saving technologies			Code
F1	Do you think household is facing any constrains that limit the adoption of WCT? 0=no 1=yes
F2	What type of constraints does household face?	1. Monetary 0=no 1=yes
		2. Information 0=no 1=yes
		3. Space/plot size 0=no 1=yes
		4. Technical expertise 0=no 1=yes
		5. Other (_____________)
F3	Which technology would you have preferred?	1. Simple rainwater tank 0=no 1=yes
		2. Simple graywater tank 0=no 1=yes
		3. Complete rainwater system 0=no 1=yes
		4. Complete graywater system 0=no 1=yes
		5. Borehole
		6. Well point
		7. Pond 0=no 1=yes

Thank you for participating in this survey

Appendix B. Testing of IVs

To be valid, the IVs must meet the 3 conditions for IVs described in Section 4.2. The first stage of the 2SLS model is easily testable and is used to determine whether an instrument meets the inclusion restriction. For this, we run an F-test to determine if the selected instruments (Z) affect the endogenous variable (X). The exclusion restriction, on the other hand, cannot be formally tested. This is because the relationship between X and Y is confounded by some error or unobservable factors, $\eta i$. Testing for conditional independence between Z and Y, while controlling for X, would therefore be confounded by the same error or unobservable factors. One is therefore required to make a strong argument for the exclusion restriction and should generally use ones which have already been proven valid in the literature. After much research and testing, we identified three appropriate IVs from the literature. For the first two, we follow Wichman et al. (2016) and many others and use fixed service fees and marginal prices (i.e., price for each consumption block) as these instruments are uncorrelated with water consumption but highly correlated with average price (Nieswiadomy and Molina 1989; Reynaud et al. 2005; Olmstead 2009; Price et al. 2014; Pérez-Urdiales and García-Valiñas 2016). We obtain the fixed fees from the Mbombela municipality water tariff schedules for 2018 and 2019. These schedules also contain the marginal prices households should pay for water in different consumption blocks and in different areas. We, however, calculated our own marginal prices for different consumption blocks (see Appendix E). To do this, we identified consumption blocks in the data and then determined the average marginal price paid by households for water in these consumption blocks. We note that this is a limitation of the study, however, it is the best option given the context of our study area. For our third IV, we follow Nauges and van den Berg (2009) and choose household location dummies as identifying instruments for the water price paid by the household. We have nine such ward dummies. These dummies are good instruments as long as household’s location choice does not depend on the price of water. We believe this to be a reasonable assumption for this study area as it is highly unlikely that household’s decision to live somewhere would be based on differences in water price. Two problems faced when using IV’s require further testing. First, the strength of the IV’s must be tested. Conducting the appropriate test showed that the instruments were jointly significant and different from zero, with an F-statistic and the Wald Test showing the instruments were strong. Second, overidentification must be tested for when more than one instrument is used. To test for overidentification, we ran the Sargan and Basman Test and Hansen Test for overidentification, which revealed the instruments were valid and not overidentified.

Appendix C. Tables and Figures

Table C.1. Description of Variables Used in Regressions

Variable	Definition
Tap Water Consumption	Monthly tap water consumption (liters)
Tap Water Price	Monthly tap water price (Rands/liter)
Food Price	Monthly food CPI
Energy Price	Monthly energy CPI
Household Income	Household income (Rands)
Age	Age of household head
Gender	Gender of household head
Education	Household head’s years of education attained
Household Size	Number of household members
Groundwater Access	Indicates groundwater access (borehole or well)
Lawn/Garden	Household owns a lawn or garden
Swimming Pool	Household owns a swimming pool
Aware of WS Strategies	Household is aware of water saving strategies
Prioritize Conservation	Household believes water conservation is a priority
Water Saving Technologies	Household installed water saving technology
Used Graywater	Household uses graywater technology
Used Rainwater	Household uses rainwater technology
Efficient Washing Machine	Household owns an efficient washing machine
Water Scarcity	Average water scarcity rating by households in a ward (0–10)
Interruptions	Number of water interruptions in the last 12 months
Water Quality	Average water quality rating by households in a ward (0–10)
Fixed Fee	Amount paid for connection to the municipal water supply system (Rands)
Marginal Price	The marginal price paid for water in different consumption blocks (R/L)

Table C.2. 2SLS Regression Results: Overall and by Income Group: Determinants of Household Tap Water Demand

		Income
Independent Variable: Water Consumption	Overall	Poor	Lower Middle	Upper Middle	Rich
Water Price	−0.672***	−0.935***	−0.235	−0.676***	−0.316
	(0.0820)	(0.137)	(0.227)	(0.254)	(0.431)
Food Price	−0.174***	−0.211*	−0.307***	−0.245***	−0.0122
	(0.0590)	(0.115)	(0.0936)	(0.0767)	(0.0959)
Energy Price	−0.0720***	−0.163***	−0.0518	−0.0817**	−0.0387
	(0.0276)	(0.0470)	(0.0355)	(0.0354)	(0.0293)
Income	0.194***	0.153*	−0.236	0.188	0.474**
	(0.0294)	(0.0911)	(0.485)	(0.491)	(0.199)
Education	0.000884	−0.00392	0.0145	0.0122	−0.0177
	(0.00476)	(0.0119)	(0.0128)	(0.0114)	(0.0158)
HH Size	0.0456***	0.0288	−0.0291	0.0714	0.0622
	(0.0133)	(0.0484)	(0.0419)	(0.0487)	(0.0633)
Swimming Pool	0.222***	0.224	0.202*	0.213	0.194
	(0.0489)	(0.301)	(0.118)	(0.138)	(0.175)
Trees/Flowers/Garden	0.139**	0.236	0.226	0.194*	−0.113
	(0.0668)	(0.191)	(0.198)	(0.113)	(0.242)
Rainwater Tank	−0.216***	0.241	−0.379**	−0.169	−0.314***
	(0.0387)	(0.186)	(0.157)	(0.144)	(0.0936)
Rainwater System	−0.329***	−0.0585	−0.534***	−0.302**	−0.346**
	(0.0588)	(0.180)	(0.181)	(0.136)	(0.154)
Efficient Washing Machine	−0.231***	−0.187	−0.225*	−0.169	−0.181
	(0.0410)	(0.153)	(0.116)	(0.118)	(0.116)
Water Quality Rating	−0.0768***	0.0139	−0.227***	−0.0807	−0.0478
	(0.0123)	(0.0406)	(0.0547)	(0.0607)	(0.0450)
Constant	33.68***	48.42***	51.88***	42.63***	9.063
	(8.573)	(17.45)	(14.07)	(12.53)	(12.61)
Observations	1,416	354	354	354	354
Wald Chi2(12)	337.88	97.79	72.20	74.49	34.98
Prob>Chi2	0.0000	0.0000	0.0000	0.0000	0.0000
R-squared	0.3992	0.5083	0.3728	0.3251	0.3722

Notes: Standard errors in parentheses

***$p<0.01$, **$p<0.05$, *$p<0.1$

Table C.3. 2SLS Regression Results by Water Sources Used: Determinants of Household Tap Water Demand

	Water Sources
Independent Variable: Water Consumption	Tap	Tap and Other	Tap and Rain	Tap and Ground	Tap and Bottled
Water Price	−0.370	−0.597***	−0.543**	−0.178	−0.794***
	(1.500)	(0.0997)	(0.268)	(0.176)	(0.112)
Food Price	−0.0220	−0.199***	−0.0363	−0.403***	−0.269***
	(0.724)	(0.0386)	(0.0684)	(0.0996)	(0.0608)
Energy Price	−0.0742	−0.0737***	−0.0280	−0.0974***	−0.105***
	(0.0899)	(0.0171)	(0.0285)	(0.0372)	(0.0255)
Income	−0.0306	0.174***	0.203**	−0.0432	0.270***
	(0.406)	(0.0451)	(0.0802)	(0.143)	(0.0846)
Education	0.0170	0.00154	−0.0179*	0.0428**	0.00870
	(0.0244)	(0.00653)	(0.00947)	(0.0214)	(0.0110)
HH Size	0.236***	0.0286	0.0240	0.0267	0.0412
	(0.0780)	(0.0239)	(0.0498)	(0.0351)	(0.0381)
Swimming Pool	−0.129	0.237***	0.0750	0.111	0.302**
	(0.193)	(0.0717)	(0.135)	(0.131)	(0.132)
Trees/Flowers/Garden	0.103	0.131	0.0985	0.245	−0.00170
	(0.300)	(0.0874)	(0.180)	(0.334)	(0.109)
Rainwater Tank	—	−0.213***	—	−0.169	—
		(0.0723)		(0.222)
Rainwater System	—	−0.350***	—	−0.498**	—
		(0.0880)		(0.252)
Efficient Washing Machine	0.297	−0.228***	−0.120	−0.215	−0.293***
	(0.833)	(0.0600)	(0.0973)	(0.168)	(0.106)
Water Quality Rating	−0.174	−0.0979***	−0.0631*	−0.327***	−0.0504
	(0.389)	(0.0301)	(0.0374)	(0.0605)	(0.0622)
Constant	19.21	37.21***	12.78	67.03***	47.24***
	(86.37)	(5.731)	(9.959)	(14.65)	(9.041)
Observations	100	1,308	625	284	446
Wald Chi2(12)	130.02	150.21	39.03	104.07	166.61
Prob>Chi2	0.0000	0.0000	0.0001	0.0000	0.0000
R-squared	0.5220	0.3980	0.2265	0.5270	0.4312

Notes: Standard errors in parentheses

***$p<0.01$, **$p<0.05$, *$p<0.1$

Table C.4. Water Price and Income Elasticities of Household Water Demand from the Developing Country Literature

Authors and Date	Country	Method	Water Price Elasticity	Income Elasticity
Rizaiza (1991)	Saudi Arabia	OLS	−0.78
Crane (1994)	Indonesia	OLS	−0.48 to −0.6	0.09 to 0.2
David and Inocencio (1998)	Philippines	2SLS	−2.1	0.3
Strand and Walker (2005)	Central America	2SLS	−0.3	0.1
Nauges and Strand (2007)	Honduras	Two-step Heckman	−0.4 to −0.7	0.2 to 0.3
Basani et al. (2008)	Cambodia	Two-step Heckman	−0.4 to −0.5	—
Cheesman et al. (2008)	Vietnam	Two-step Heckman	−0.06 to −0.53	—
Nauges and van den Berg (2009)	Sri Lanka	Two-step Heckman	−0.15 to −0.37	0.14
Nauges and Whittington (2010)	Developing Countries	Meta-analysis	−0.3 to −0.7	0.1 to 0.3
Miyawaki et al. (2011)	Japan	Tobit Model	−0.37 to −0.51	0.14 to 0.57
Coulibaly et al. (2014)	Jordan	MLR	−1.33	—
Kotagama et al. (2017)	Oman	2SLS	−2.1	—
Ahmad et al. (2017)	Pakistan	2SLS	−0.43 to −0.7	0.01 to 0.12
Taştan (2018)	Turkey	ARDL Framework	−0.15 to −.0.3	—
Zhu et al. (2018)	Global	Meta-analysis	−0.23 to −0.58	0.24 to 0.96
Abdullah et al. (2019)	Nigeria	OLS, 2SLS, 3SLS	—	0.268
Wagner et al. (2019)	Kenya	Probit	−0.13 to −1.33 (−0.56 mean)	—
Gross and Elshiewy (2019)	Benin	CMNL	−0.26	—
PalancaTan (2020)	Philippines	OLS	−0.38	—
Ščasný and Smutná (2021)	Czech Republic	2SLS	−0.22 to −0.3	0.16
Oyerinde and Jacobs (2022)	Nigeria	OLS		0.093

Table C.5. Descriptive Statistics by Water Sources Used

		Water Sources
	Overall	Tap	Tap and Other	Tap and Rain	Tap and Ground	Tap and Bottled
Tap water consumption (lpm)	2,268	1,729	2,306	1,778	3,806	2,114
Average water bill (Rpm)	918	825	929	807	1,272	927
Socio-Economic
Income	26,878	16,343	27,662	29,785	26,218	25,426
Gender ( HH Head Male)	68%	66%	68%	68%	67%	68%
Age of HH head	53.3	54.0	53.3	53.9	53.0	52.9
No. of employed HH members	1.4	1.1	1.5	1.7	1.2	1.2
HH years of education	14.7	10.4	15.0	15.1	15.5	14.7
Spouse tears of education	9.9	7.7	10.0	10.6	9.7	9.3
HH size	4.0	4.0	4.0	4.0	4.1	3.9
Children under 5	0.4	0.6	0.4	0.2	0.5	0.5
Housing Characteristics
Owns trees/flowers/garden	90%	76%	91%	93%	93%	88%
Owns swimming pool	22%	12%	23%	12%	42%	29%
Water Saving Technologies
Uses graywater technologies	64%	42%	66%	76%	64%	54%
Uses rainwater technologies	69%		72%	97%	53%	52%
Uses efficient shower head	50%	57%	50%	49%	52%	51%
Owns efficient washing machine	39%	43%	39%	48%	27%	33%
Uses efficient toilet	50%	40%	51%	64%	47%	39%
Aware of WS strategies	92%	84%	93%	100%	96%	84%
Prioritize water conservation	93%	80%	94%	97%	97%	89%
Water Supply Characteristics
Water scarcity rating (/10)	8.6	8.5	8.6	8.0	9.0	8.9
No. of interruptions	1.7	2.2	1.7	2.1	1.6	1.2
Water quality rating (/10)	8.8	9.0	8.8	8.4	8.9	9.2
Number of Observations	472	33	436	208	95	149

Table C.6. OLS Regression Results: Overall and by Income Group: Determinants of Household Tap Water Demand

		Income
Independent Variable: Water Consumption	Overall	Poor	Lower Middle	Upper Middle	Rich
Water Price	−0.526***	−0.392***	−0.678***	−0.489***	−0.519***
	(0.0513)	(0.0788)	(0.0834)	(0.131)	(0.102)
Food Price	−0.190***	−0.259***	−0.243***	−0.256***	0.0114
	(0.0397)	(0.0886)	(0.0730)	(0.0723)	(0.0773)
Energy Price	−0.0738***	−0.136***	−0.0643*	−0.0777**	−0.0151
	(0.0163)	(0.0371)	(0.0349)	(0.0320)	(0.0265)
Income	0.117**	0.0527	−0.227	0.0777	0.448**
	(0.0468)	(0.115)	(0.387)	(0.497)	(0.201)
Education	0.00244	0.0101	0.0103	0.0177	−0.0227*
	(0.00613)	(0.0128)	(0.0116)	(0.0123)	(0.0134)
HH Size	0.0389*	0.0260	0.0122	0.0661	0.0733
	(0.0210)	(0.0455)	(0.0327)	(0.0455)	(0.0458)
Swimming Pool	0.134**	−0.118	0.130	0.155	0.0964
	(0.0659)	(0.221)	(0.100)	(0.139)	(0.147)
Trees/Flowers/Garden	0.156*	0.368*	0.0623	0.238*	−0.127
	(0.0831)	(0.211)	(0.150)	(0.130)	(0.222)
Rainwater Tank	−0.224***	−0.0757	−0.391***	−0.125	−0.321***
	(0.0567)	(0.159)	(0.148)	(0.136)	(0.0668)
Rainwater System	−0.345***	−0.248	−0.397***	−0.318**	−0.367***
	(0.0817)	(0.206)	(0.153)	(0.128)	(0.135)
Efficient Washing Machine	−0.235***	−0.363**	−0.190*	−0.154	−0.203*
	(0.0571)	(0.143)	(0.114)	(0.115)	(0.110)
Water Quality Rating	−0.0915***	−0.0572	−0.221***	−0.0791	−0.0608
	(0.0275)	(0.0490)	(0.0466)	(0.0598)	(0.0423)
Constant	36.50***	52.16***	45.44***	44.24***	3.439
	(5.808)	(13.50)	(12.31)	(12.05)	(10.18)
	1,416	354	354	354	354
${WaldChi}^2(12)$	206.60	76.66	151.28	71.73	96.68
$Prob>{Chi}^2$	0.0000	0.0000	0.0000	0.0000	0.0000
R-squared	0.3951	0.5046	0.5138	0.3375	0.4288

Notes: Standard errors in parentheses

***$p<0.01$, **$p<0.05$, *$p<0.1$

Table C.7. OLS Regression Results by Water Sources Used: Determinants of Household Tap Water Demand

	Water Sources
Independent Variable: Water Consumption	Tap	Tap and Other	Tap and Rain	Tap and Ground	Tap and Bottled
Water Price	−0.229	−0.550***	−0.454***	−0.637***	−0.571***
	(0.235)	(0.0526)	(0.0862)	(0.130)	(0.0725)
Food Price	−0.134	−0.201***	−0.0373	−0.417***	−0.298***
	(0.169)	(0.0386)	(0.0615)	(0.0800)	(0.0561)
Energy Price	−0.0834	−0.0736***	−0.0330	−0.125***	−0.102***
	(0.0832)	(0.0164)	(0.0265)	(0.0324)	(0.0241)
Income	−0.133	0.110**	0.158**	0.166	0.172*
	(0.179)	(0.0482)	(0.0669)	(0.153)	(0.0891)
Education	0.0195	0.00209	−0.0171*	0.0402**	0.0154
	(0.0272)	(0.00638)	(0.00974)	(0.0178)	(0.0113)
HH Size	0.240***	0.0275	0.0162	0.0238	0.0343
	(0.0687)	(0.0221)	(0.0392)	(0.0278)	(0.0358)
Swimming Pool	−0.147	0.159**	0.00999	0.0623	0.231*
	(0.116)	(0.0689)	(0.116)	(0.115)	(0.131)
Trees/Flowers/Garden	0.0711	0.137	0.111	0.147	0.0670
	(0.192)	(0.0927)	(0.180)	(0.318)	(0.115)
Rainwater Tank	—	−0.212***	0.208	—	—
		(0.0704)	(0.281)
Rainwater System	—	−0.346***	0.0473	—	—
		(0.0824)	(0.272)
Efficient Washing Machine	0.357	−0.233***	−0.139	−0.243	−0.293***
	(0.227)	(0.0594)	(0.0857)	(0.175)	(0.101)
Water Quality Rating	−0.214***	−0.106***	−0.0735*	−0.276***	−0.0692
	(0.0769)	(0.0303)	(0.0396)	(0.0546)	(0.0596)
Constant	34.05	37.95***	13.84	68.90***	51.18***
	(25.40)	(5.671)	(8.705)	(12.83)	(8.325)
Observations	100	1,308	625	228	446
${WaldChi}^2(12)$	107.84	242.00	68.49	247.07	152.50
$Prob>{Chi}^2$	0.0000	0.0000	0.0000	0.0000	0.0000
R-squared	0.5135	0.4201	0.2511	0.7356	0.4348

Notes: Standard errors in parentheses

***$p<0.01$, **$p<0.05$, *$p<0.1$

Table C.8. 2SLS Regression, OLS Regression, and FEs Model Results: Determinants of Household Tap Water Demand

	Regression Model
Independent Variable: Water Consumption	2SLS	OLS	FEs
Water Price	−0.672***	−0.526***	−0.615***
	(0.0820)	(0.0513)	(0.0214)
Food Price	−0.174***	−0.190***	−0.331**
	(0.0590)	(0.0397)	(0.133)
Energy Price	−0.0720***	−0.0738***	−0.299***
	(0.0276)	(0.0163)	(0.0759)
Income	0.194***	0.117**	0.151***
	(0.0294)	(0.0468)	(0.0272)
Education	0.000884	0.00244	0.00924**
	(0.00476)	(0.00613)	(0.00369)
HH Size	0.0456***	0.0389*	0.0344***
	(0.0133)	(0.0210)	(0.0111)
Swimming Pool	0.222***	0.134**	0.143***
	(0.0489)	(0.0659)	(0.0368)
Trees/Flowers/Garden	0.139**	0.156*	0.135***
	(0.0668)	(0.0831)	(0.0505)
Rainwater Tank	−0.216***	−0.224***	−0.190***
	(0.0387)	(0.0567)	(0.0349)
Rainwater System	−0.329***	−0.345***	−0.276***
	(0.0588)	(0.0817)	(0.0464)
Efficient Washing Machine	−0.231***	−0.235***	−0.160***
	(0.0410)	(0.0571)	(0.0316)
Water Quality Rating	−0.0768***	−0.0915***	−0.0267*
	(0.0123)	(0.0275)	(0.0154)
Constant	33.68***	36.50***	79.64***
	(8.573)	(5.808)	(19.85)
Observations	1,416	1,416	1,416
${WaldChi}^2(12)∕$F-Stat	337.88	206.60	55.07
$Prob>{Chi}^2∕Prob>F$	0.0000	0.0000	0.0000
R-squared	0.3992	0.3951	0.5252

Table C.9. FEs Regression Results: Overall and by Income Group: Determinants of Household Tap Water Demand

		Income
Independent Variable: Water Consumption	Overall	Poor	Lower Middle	Upper Middle	Rich
Water Price	−0.615***	−0.616***	−0.631***	−0.727***	−0.573***
	(0.0214)	(0.0389)	(0.0493)	(0.0680)	(0.0518)
Food Price	−0.331**	−0.502*	−0.755***	−0.120	−0.211
	(0.133)	(0.271)	(0.277)	(0.283)	(0.325)
Energy Price	−0.299***	−0.212	−0.399***	−0.234	−0.655**
	(0.0759)	(0.147)	(0.143)	(0.164)	(0.289)
Income	0.151***	0.101	−0.312	0.0808	0.381***
	(0.0272)	(0.0663)	(0.228)	(0.261)	(0.106)
Education	0.00924**	0.00571	0.0139*	0.0269***	0.000410
	(0.00369)	(0.00767)	(0.00749)	(0.00787)	(0.00776)
HH Size	0.0344***	0.0299	0.0154	0.0655***	0.0395
	(0.0111)	(0.0279)	(0.0183)	(0.0236)	(0.0245)
Swimming Pool	0.143***	−0.155	0.199***	0.230***	0.0365
	(0.0368)	(0.167)	(0.0546)	(0.0682)	(0.0823)
Trees/Flowers/Garden	0.135***	0.235*	0.0796	0.303***	−0.139
	(0.0505)	(0.122)	(0.0791)	(0.112)	(0.104)
Rainwater Tank	−0.190***	0.0997	−0.240**	−0.145*	−0.300***
	(0.0349)	(0.0990)	(0.0996)	(0.0777)	(0.0547)
Rainwater System	−0.276***	−0.203	−0.345***	−0.145	−0.363***
	(0.0464)	(0.127)	(0.0881)	(0.0881)	(0.0719)
Efficient Washing Machine	−0.160***	−0.342***	−0.177**	−0.0477	−0.0465
	(0.0316)	(0.0853)	(0.0692)	(0.0675)	(0.0586)
Water Quality Rating	−0.0267*	−0.0187	−0.144***	−0.00828	0.0234
	(0.0154)	(0.0265)	(0.0355)	(0.0256)	(0.0248)
Constant	79.64***	87.86**	144.9***	48.11	108.8**
	(19.85)	(40.69)	(41.65)	(40.41)	(48.22)
Observations	1,416	354	354	354	354
F-Stat	55.07	31.95	20.41	39.23	22.41
$Prob>F$	0.0000	0.0000	0.0000	0.0000	0.0000
R-squared	0.5252	0.5610	0.5919	0.5173	0.5683

Notes: Standard errors in parentheses

***$p<0.01$, **$p<0.05$, *$p<0.1$

Table C.10. FEs Regression Results by Water Sources Used: Determinants of Household Tap Water Demand

	Water Sources
Independent Variable: Water Consumption	Tap	Tap and Other	Tap and Rain	Tap and Ground	Tap and Bottled
Water Price	−0.499***	−0.602***	−0.521***	−0.649***	−0.577***
	(0.183)	(0.0220)	(0.0481)	(0.0376)	(0.0400)
Food Price	−0.155	−0.332**	−0.423*	−0.680*	0.0242
	(0.436)	(0.140)	(0.217)	(0.387)	(0.197)
Energy Price	−0.0534	−0.299***	−0.167	−0.740***	−0.189**
	(0.226)	(0.0792)	(0.156)	(0.149)	(0.0892)
Income	−0.0494	0.157***	0.179***	0.0251	0.258***
	(0.147)	(0.0275)	(0.0392)	(0.0719)	(0.0557)
Education	0.0245	0.0103**	−0.00955	0.0433***	0.0218***
	(0.0175)	(0.00404)	(0.00595)	(0.00914)	(0.00668)
HH Size	0.234***	0.0223*	0.0195	0.0147	0.0167
	(0.0468)	(0.0118)	(0.0219)	(0.0147)	(0.0204)
Swimming Pool	−0.0531	0.157***	−0.0513	0.0896	0.203***
	(0.118)	(0.0392)	(0.0731)	(0.0675)	(0.0723)
Trees/Flowers/Garden	0.192	0.0933*	0.192*	−0.0474	−0.00252
	(0.149)	(0.0522)	(0.109)	(0.102)	(0.0636)
Rainwater Tank	—	−0.180***	—	—	—
		(0.0395)
Rainwater System	—	−0.275***	—	—	—
		(0.0472)
Efficient Washing Machine	0.110	−0.145***	−0.0207	−0.192**	−0.236***
	(0.145)	(0.0331)	(0.0483)	(0.0803)	(0.0581)
Water Quality Rating	−0.148***	−0.0273	−0.0287	−0.0930**	0.0173
	(0.0538)	(0.0175)	(0.0233)	(0.0381)	(0.0338)
Constant	31.12	79.91***	72.94**	175.9***	24.88
	(61.85)	(20.75)	(32.06)	(51.50)	(28.00)
Observations	100	1,308	625	228	446
F-Stat	16.08	76.98	15.73	44.27	31.28
$Prob>F$	0.0000	0.0000	0.0000	0.0000	0.0000
$AdjustedR$-squared	0.5707	0.5228	0.3601	0.8365	0.5305

Notes: Standard errors in parentheses

***$p<0.01$, **$p<0.05$, *$p<0.1$

Table C.11. Sensitivity Analysis Results: 2SLS Regression Results of the Determinants of Household Water Demand (New Water Price Variable)

		Income
Independent Variable: Water Consumption	Overall	Poor	Lower Middle	Upper Middle	Rich
Water Price (No Fixed Fee)	−0.460***	−0.844***	−0.0337	−0.467*	0.0344
	(0.116)	(0.190)	(0.211)	(0.266)	(0.380)
Food Price	−0.225***	−0.431**	−0.342***	−0.249**	0.0555
	(0.0468)	(0.168)	(0.114)	(0.101)	(0.105)
Energy Price	−0.0885***	−0.286***	−0.0400	−0.0717	−0.0134
	(0.0231)	(0.0775)	(0.0439)	(0.0454)	(0.0367)
Income	0.222***	0.187	−0.351	0.204	0.477**
	(0.0533)	(0.146)	(0.529)	(0.557)	(0.216)
Education	0.00333	−0.00554	0.0160	0.0139	−0.0120
	(0.00813)	(0.0178)	(0.0136)	(0.0129)	(0.0175)
HH Size	0.0467**	0.00541	−0.0376	0.0757	0.0398
	(0.0238)	(0.0659)	(0.0475)	(0.0562)	(0.0749)
Swimming Pool	0.225***	0.298	0.178	0.251*	0.233
	(0.0848)	(0.421)	(0.137)	(0.152)	(0.188)
Trees/Flowers/Garden	0.182	0.469	0.267	0.238*	−0.170
	(0.114)	(0.336)	(0.208)	(0.140)	(0.330)
Rainwater Tank	−0.285***	0.252	−0.331*	−0.202	−0.332***
	(0.0652)	(0.259)	(0.180)	(0.170)	(0.125)
Rainwater System	−0.399***	−0.0592	−0.510***	−0.316**	−0.304
	(0.0996)	(0.260)	(0.189)	(0.160)	(0.192)
Efficient Washing Machine	−0.262***	−0.211	−0.190	−0.218	−0.180
	(0.0707)	(0.216)	(0.133)	(0.137)	(0.134)
Water Quality Rating	−0.0863***	0.0314	−0.232***	−0.0818	−0.0274
	(0.0210)	(0.0627)	(0.0557)	(0.0679)	(0.0459)
Constant	41.17***	87.40***	55.51***	41.63**	−1.492
	(6.865)	(25.89)	(17.15)	(16.32)	(14.11)
Observations	1,416	354	354	354	354
${WaldChi}^2(12)$	128.06	41.04	39.90	49.07	25.57
$Prob>{Chi}^2$	0.0000	0.0000	0.0001	0.0000	0.0123

Notes: Standard errors in parentheses

***$p<0.01$, **$p<0.05$, *$p<0.1$

Table C.12. Scenario Descriptions and Methodologies

Period	Scenario	Description	Methodology/Calculations
2019	FOOM	There is a boom in food prices, while everything else remains the same.	We implement a 53% increase in food prices, while everything else remains the same. For the increase value, we take an average of the food price increases in the three most recent food price booms: (1) the 2007/2008 boom ($+$69%), (2) the 2010/2011 boom ($+$46%), and the COVID boom ($+$43%). The data is obtained from the Food and Agricultural Organization (FAO 2023) Food Price Index.
	FREEW	Free water and electricity are provided to all (up to limited thresholds).	We presume that water and electricity is provided free, and hence implement a 100% decrease in the price of water and energy. We include a free water threshold of 6,000lpm, and for electricity it is 50kWh/pm. To note is that we have focused on energy prices and not electricity prices specifically in the analysis. The reason we now focus on electricity only because it is highly unlikely that the free provision of other energy sources will be possible in South Africa. There is potential that the price of other energy sources could still impact water consumption, and we note this is a limitation of this scenario.
2035	BAU 2035	BAU trends in WEF prices, as well as income, continue to 2035.	We assume BAU increases in the price of WEF as well as income for 16 years (2019–2035). We obtain the price increase data from Stats SA (2013–2023) and estimate annual increases equal to the average from the last 10 years (water: $+$2.58% p.a., energy: $+$4.43% p.a., food: $+$4.31% p.a.). We obtain income data from the Stats SA (2018–2024) Income and Expenditure Survey and estimate annual increases equal to the average of the last 5 years ($+$4.88% p.a.).
	CC 2035	WEF prices increase due to climate change impacts by 2035.	We implement predicted increases in WEF prices due to climate change by 2035. For food, we implement a 77.6% increase in price. This is because recent work has shown that food prices will likely increase by 50% from 2023 to 2035 (European Central Bank 2023). We thus first calculate food price increases from 2019 to 2023 following BAU trends (as described above) and then use this as a base for the 50% increase. For water, there is limited research on how prices will be affected by climate change, although we do know that it will drive severe water scarcity in South Africa in the future, with large water deficits predicted (WWF-SA 2016). We thus assume the increased water scarcity will result in a price effect at least the magnitude of the food price increases ($+$50% from 2023). Following the same procedure as was done for the food price calculations, the water price increase from 2019 to 2035 is 66%. For energy and income, we assume that they are not significantly affected and follow the BAU 2035 trends ($+$4.43% p.a. for energy and $+$4.88% p.a. for income).

Appendix D. Data Cleaning

Following data collection, we undertook a data cleaning process to prevent biases and inconsistencies caused by data capture errors. During this, we removed observations with incorrect or misreported data entries. For example, observations with negative values for water consumption, water bills, etc. Researchers are often tempted to remove outliers from a sample, but they can contain important information about the study area, particularly in unequal settings like our study area. Outliers should thus only be removed when there is a clear measurement error. We focused on identifying outliers with respect to the independent variable in our study (household water consumption). Given the level of inequality that characterizes in our study area, outliers in terms of water consumption were expected and we thus only remove observations for which the water consumption values were more or less than 10 times the average. This is because previous studies in similar contexts have shown household water consumption levels with ranges of similar magnitude (Dang et al. 2022; Grespan et al. 2022). This process only resulted in the removal of three households from the sample. Removing observations with measurement error and outliers resulted in dropping a total of 52 households from the sample.

Appendix E. Issues with Tap Water Price in the Study Area

The tap water prices paid by households in our study area were heavily swayed by corrupt practices (Viljoen 2019; Van Rensburg 2019). The municipal water tariff schedules for Mpumalanga during the study period state that the first 6,000L of municipal water should be provided free, except for the fixed fee that must be paid to be connected to the municipal water supply system. The average tap water consumption level of households in our study area was 2,242 lpm, with only 8% of households consuming more than the 6,000lpm threshold for free water. Despite this, the average water bill received by households was R 917.87 per month and for households consuming less than 6,000lpm it was R 761.34 per month. The fixed fees during the study period ranged from R 118 to R 125 per month. It is thus clear that households were not receiving their first 6,000L free and instead paying large amounts per liter. For households consuming less than 6,000lpm, the average price paid for water was R 0.42 per liter, which is nearly 13 times higher than the marginal price per liter in the highest consumption block reported in the tariff schedule (R 0.03 per liter). The prices paid for water also varied widely by ward. Given this finding, we investigated what was happening with water bills in the study area at the time. The evidence found showed that residents were reporting receiving water bills up to 11 times higher their usual amount (Viljoen 2019; Van Rensburg 2019). The main reason we found for this was that wards in Mpumalanga were in substantial debt, with municipalities in the region owing more than R 1 billion for water (Coetzee 2020). To reduce this debt, they were charging illegal prices for municipal to try increase their revenue. Another reason for over pricing was due to illegal water connections which meant wards would charge households with legal connections more to compensate for the non-payment on illegal connections (Coetzee 2020). Given this, we had to calculate the tap water price paid by households in our study area and could not use the tariff schedule. We hence followed Deaton (1990) and calculated the unit values for tap water by ward and consumption block (see Section  3).

ORCID

Thomas van Huyssteen  https://orcid.org/0000-0001-8494-8269

Djiby Thiam  https://orcid.org/0000-0001-5971-4030

Sanderine Nonhebel  https://orcid.org/0000-0002-6625-0745