Analysis of physical and non-physical factors associated with individual water consumption using a hierarchical linear model before and after an earthquake in a region with insufficient water supply

The quantity of domestic water influences hygiene and therefore public health (Howard & Bartram 2003). The World Health Organization categorised households consuming less than 20 litre/capita/day (LPCD) as ‘basic access’, those consuming 50 LPCD as ‘intermediate access’ and those consuming 100 LPCD or more as ‘optimal access’ (Howard & Bartram 2003). However, there are many regions where individual water consumption (LPCD) does not reach ‘optimal access’ (Shaban & Sharma 2007; Fan et al. 2013; Otaki et al. 2022). According to the United Nations (Sustainable Development Goals Goal 6), water scarcity affects more than 40% of the global population and is expected to increase. Recently, demand-side water management has attracted global attention, and several reports have focused on water use in households (Zerah 2000; Pattanayak et al. 2005; Fan et al. 2013; Grace et al. 2013; Pasakhala et al. 2013; Guragai et al. 2017). Achore et al. (2020) identified that nine key coping strategies are typically employed by households, including constructing alternative water sources, buying water from private vendors, water storage, water sharing and borrowing from social networks, and treating water to deal with water quality and quantity issues. In addition, households actively collaborate to facilitate water treatment, distribution and protection through community participation in regions with water scarcity (Shrestha et al. 2019). Ito et al. (2021) defined factors which vary at the household level, including coping strategies, socioeconomic status and community participation, as ‘non-physical’ factors.

The Kathmandu Valley, Nepal, is one region suffering from chronic water scarcity. Water demand is 472 million litres per day (MLD), whereas Kathmandu Upatyaka Khanepani Limited (KUKL), which is the only piped water supply company for the region, supplies 97 and 126 MLD during the months of minimum and maximum production, respectively (KUKL Annual Report 2021). Though most existing research on water consumption at household have not focused on coping strategies and community participation (Babel et al. 2007; Nauges & van den Berg 2009; House-Peters & Chang 2011; Coulibaly et al. 2014), Ito et al. (2021) attempted to quantitatively evaluate the factors associating with water consumption not only from the perspective of physical factors which depend on the external environment but also from the perspective of non-physical factors in the Kathmandu Valley. In other words, water source and supply time (i.e. physical factor), and wealth status, education for household head, house ownership, participation in local community and water treatment (i.e. non-physical factors) were selected to assess their association with individual water consumption. Ito et al. (2021) showed that the number of water sources was associated with increased water consumption and water treatment was associated with decreased water consumption regardless of the impact of Gorkha earthquake and seasonality. Wealth status, education for household heads, house ownership and participation in the local community were associated with increased water consumption before the earthquake during the dry season but these associations were not apparent after the earthquake during dry season. In the Valley, characteristics such as water quality and supply quantity may differ in each KUKL water supply area, and the regional disparity of waterborne infection risk from multiple water sources was indicated (Ito et al. 2020). In addition, location-specific factors such as culture and location may influence the relationship between water consumption and each factor. Therefore, supply areas should be strongly considered to indicate the factors associated with individual water consumption, especially in the context of regions with insufficient water supply. The inclusion of the ‘water supply area’ factor as one of the independent variables in a linear regression model was attempted, but its impact was too large to confirm the impact of the other variables. Behavioural and social data commonly have a nested structure (Raudenbush & Bryk 2002), and the hierarchical linear regression model (HLM) is recommended to address issues associated with the hierarchical nature of multilevel data (Hofmann 1997; Raudenbush & Bryk 2002; Kim et al. 2009). This approach has been extensively used in education and psychology but is rarely used in the field of water sciences. In this study, households were considered to be nested within the water supply area, that is, the water supply area had an indirect influence on the association between individual water consumption and each factor. This study applies the HLM to build on the model developed by Ito et al. (2021) by considering the effect of water supply area and to analyse nested data. Furthermore, we used the HLM model to confirm the physical and non-physical factors associated with individual water consumption (LPCD) throughout the Kathmandu Valley.

On 25 April 2015, a 7.6 magnitude earthquake (hereinafter called ‘Gorkha earthquake’) struck Barpak in the historic district of Gorkha, about 76 km northwest of Kathmandu. Gorkha earthquake caused major (partial) damage to 1,570 (3,663) out of 11,288 water supply systems in the affected areas (NPC 2015). Thapa et al. (2016) indicated an almost 40% reduction in piped water supply in the Kathmandu Valley after the earthquake.

Table 1 depicts the dependent and independent variables selected in this study for HLM.

In the Kathmandu Valley, the main water sources are piped water, groundwater, tanker water and jar water. Groundwater recharge is higher during the wet season (1.949 million-cubic metres: MCM) than during the dry season (1.052 MCM) (Lamichhane & Shakya 2020). The unit cost for using groundwater (0.18 USD/cubic metre) is estimated to be significantly cheaper than that for using tanker water (2.22 USD/cubic metre) (Ojha et al. 2018). Some households use rainwater, neighbour's piped water, neighbour's well water, river/lake/pond water, public well water, spring water and stone spout water. Interviewers asked households about their water consumption (in L) per day for each source separately, referring to the size of containers habitually used in the household, and the amount divided by the number of people in the household was determined as the water consumption LPCD. The sum of water consumption from different sources was considered as the total water consumption, a dependent variable.

Independent variables were established for both physical and non-physical characteristics based on Ito et al. (2021). As physical factors, ‘source’ and ‘supply time’ were identified. ‘Source’ consisted of four categories: households using only piped water, groundwater (and piped water), tanker water (and piped water) and groundwater and tanker water (and piped water). ‘Supply time’ was the hours per week that piped water was supplied. In addition, ‘No. of families’, ‘Wealth status’, ‘Head education’, ‘Ownership’, ‘Treatment’ and ‘Community’ were identified as non-physical factors. The number of members in the respondent's family was indicated by ‘No. of families’ and this variable was not included in Ito et al. (2021). ‘Wealth status’ was determined by constructing a wealth index based on 16 household asset possessions (e.g. television, bicycle, car and refrigerator) (Cordova 2009). The wealth index was calculated by weighing each asset identified via principal component analysis. The wealth index reflects a household's long-term economic situation. Based on the wealth index, households were categorised into five quintiles: lowest/lower/medium/higher/highest (Shrestha et al. 2017). ‘Head education’ consists of three categories based on the level of education attained by the household head: no education, primary or secondary school and college or university. ‘Ownership’ (i.e. house owner/tenant) was also considered. ‘Treatment’ consists of two categories: with/without water treatment regardless of the method of treatment. ‘Community’ consists of three categories: no participation (in the local community), no community water (household participates in the local community but does not use water stored and managed by the community), and use community water (household participates in the local community and uses water stored and managed by the community), whereas the variable in our previous report consisted of two categories. Correlations between independent variables have been confirmed to avoid the problem of multicollinearity during linear regression analysis (Supplementary Material, Table S1). Although the storage of piped water and/or groundwater (yes/no) is one of the coping strategies, it was not included as an independent variable because it was strongly correlated with ‘source’. In addition, though the caste system may be one of the explanatory variables, it was not included in this analysis because of the inequivalent sample sizes among the four categories. Instead, the ‘wealth status’ variable is included to consider characteristics of the caste system.

Data from households that did not completely answer all questions related to water consumption and the above eight were excluded from the analysis. Because only three households responded completely in Phase 1, supply area 9 was excluded from the analysis. As a result, the data collected from 732, 833 and 860 of 992 households in Phases 1, 2 and 3, respectively, were used in the analysis. The number of samples in each area is described in Table 2.

Ito et al. (2021) determined the variables for the regression analysis. However, our previous model was not able to consider the effects of water supply areas. This study builds on the model developed by Ito et al. (2021) the model by considering the random effect of water supply area using an HLM, a two-level linear model of households and supply areas, composed of the following four sub-models:

Model 1: Null model

Model 2: Random intercept model (non-random slope)

Model 3: Random slope model (non-intercept slope)

Model 4: Random intercept and random slope model

Several patterns with and without the random effect of the intercept and slopes for each explanatory variable were tested on the basis of the above models. The model with lower values of Akaike's information criterion and Schwarz's Bayesian information criterion was considered to be the model most suitable for the data.

Comparisons using the Games–Howel test or Tukey's honestly significant difference (HSD) test were used to examine the difference in water consumption of each water source among the three phases. Welch's analysis of variance (ANOVA) was performed to confirm the difference in the water consumption among the water supply areas. The correlation between continuous variables and between continuous and nominal variables was assessed using the Pearson correlation coefficient (PCC), and the correlation between nominal variables was assessed using Cramer's V. The correlation was considered strong when the PCC exceeded 0.7 (Dancey & Reidy 2007) and Cramer's V exceeded 0.5. The regression coefficient and significance levels calculated using the HLM were used to show the impact of the selected factors on individual water consumption. The significance level was set at <0.05 for all statistical analyses.

The ethical review board of the University of Yamanashi and the Nepal Health Research Council reviewed and approved the study protocol, with application number 1 (28 November 2014) and 262/2014 (18 January 2015), respectively (Shrestha et al. 2017). The participants were informed about the objectives and procedures of the study, and their voluntary participation was requested. The anonymity and confidentiality of the participants were ensured. Those who agreed to the terms and conditions signed an informed consent form. During the interview, participants could skip questions and withdraw from the study at any time.

Table 3 shows the result of the null model for each phase, and the ICC was 0.24, 0.17 and 0.21 in Phases 1, 2 and 3, respectively. All of ICC values were greater than the criterion (0.1) for all periods and it was confirmed that the undetected bias from random effects of intercept and slope described below was removed. The results of the best model for each phase are shown in Table 4.

During the baseline period, the best models had a random intercept and random slopes for source, wealth status, head education and community. Households using multiple water sources significantly consumed more water than those using only piped water. In addition, households with a higher wealth status tended to consume more water. On the other hand, households with many family members used less water per person. The other independent variables, namely, supply time, head education, ownership, treatment and community, are not associated with water consumption.

During the wet season, in Phase 3, the best models had a random intercept and random slopes for source and treatment. Compared with the baseline period, there was no significant difference between households using only piped water and tanker water (and piped water). Furthermore, supply time and ownership were strongly associated with water consumption, whereas no association between wealth status and water consumption was observed.

After the earthquake, in Phase 2, the best models had random slopes for source and treatment. Compared with the baseline period, the significant association of wealth status with water consumption disappeared, whereas that of a community with water consumption appeared. Households participating in the local community significantly consume more water, especially when the water being used was managed by the community.

The association of water consumption with water sources was consistent with the findings in Ito et al. (2021). Therefore, the results of this study emphasised that diversifying water sources is an effective strategy for obtaining water even during an emergency. Securing multiple water sources was effective for obtaining water volume because of the limited water from each source. After the earthquake, the accessibility of water sources was severely interrupted by the break of underground water supply pipes, destruction of wells and obstruction of tanker water distribution due to road collapses. However, households with multiple water sources were able to rely on another source when the availability of one water source decreased. External supports such as the construction of new wells and improvement of roads for portable water are options for policy makers to enable diversification of water sources. The random slope for supply time was not considered for all phases and the association of water consumption with supply time was consistent with our previous report (Ito et al. 2021).

The association that households with many family members consumed less water per person regardless of the supply area is consistent with previous reports (Keshavarzi et al. 2006; Fan et al. 2013). This study found that rationing water within the household was an effective water conservation strategy, even after the earthquake. Households with higher wealth status tended to consume more water during the baseline period. However, this association was not observed, and there was no difference in the association between water supply areas in Phases 2 and 3. This could be due to the sufficiency of water during the wet season, which is a common phenomenon, and the equal restrictions on market water distribution due to the earthquake in all wealth statuses. The education of the household head was not associated with water consumption, and this result is inconsistent with previous reports (Fan et al. 2013; Ito et al. 2021). Our study confirmed that house owners consumed more water only during the wet season, and this difference was due to groundwater consumption (t-test: p < 0.01). There is a possibility that house owners preferentially use groundwater and tenants can only use a limited amount of it and this gap is particularly notable during the wet season when groundwater recharge is larger. The random slope for treatment was considered only after the earthquake because the earthquake likely caused regional differences in the affordability of water treatment kits. The significant association between water consumption and treatment was not confirmed throughout the three periods – contrary to the result of Ito et al. (2021). Households typically treat water only for drinking; therefore, water treatment does not influence total water consumption.

Community involvement was significantly associated with water consumption only during an emergency (Phase 2). According to a previous report (Bisung & Elliott 2014), the interaction among community members influences their ability to collectively formulate and enforce rules for managing water and sanitation facilities. In addition, social cohesion may influence the ability to enforce and/or reinforce group or social norms for positive health behaviours (McNeill et al. 2006). During an emergency, when water supply and distribution of market water were restricted, community cohesion might have been strengthened in many communities, and residents actively shared water and water-related information. In addition, earthquake damage to stored and managed water may be quickly repaired by community members. Therefore, residents are encouraged to participate in the local community to enable using water stored and managed by the community and introducing compact water treatment systems (Shrestha et al. 2019) that are decentralised and managed by residents is one of the effective measures.

This study identified factors associated with individual water consumption in the Kathmandu Valley using data from randomly selected households and the HLM. This approach helps understand the situation in regions where water use is diverse due to intermittent water supply. Analysis of coping strategies can be improved using more detailed information about water storage such as tank type and size and frequency of water delivery/collection (e.g. via rainfall, via groundwater, via tanker, etc.). Furthermore, the impact of buying water, which was not included as a variable because less than 8% of households used only market water, can be confirmed using water-related expense data. Information about the details of the community (i.e. activity and frequency of activity) is required to evaluate the process of increasing water consumption through community involvement. In addition, the earthquake's damage to the piped water system persisted until Phase 3, as shown in Figure 3; thus, data collected during an ordinary wet season should be used to more accurately confirm the impact of the ‘seasonality’.

The physical (i.e. water source and supply time) and non-physical (i.e. number of families, wealth status, education for household head, house ownership, water treatment and community involvement) factors associated with individual water consumption throughout the Kathmandu Valley were confirmed using the HLM. The main findings of this study were: (1) households using many water sources consumed more water regardless of supply area, (2) households with many family members used less water per person, (3) during an emergency, households participating in the local community consumed more water than households not participating in the community, especially when the water being used was managed by the community.

We thank the interviewers for their hard work and the questionnaire respondents for their cooperation. We would like to thank The Small Earth Nepal (SEN) for conducting the survey. This study was partially supported by the Science and Technology Research Partnership for Sustainable Development (SATREPS) funded by Japan International Cooperation Agency (JICA)/Japan Science and Technology Agency (JST), Solution-Driven Co-creative R&D Program for SDGs (SOLVE for SDGs) funded by JST RISTEX (JPMJRX21I7), and Grants-in-Aid for Scientific Research funded by JSPS (JP21J14179).

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

The authors declare there is no conflict.