Clean water, sanitation and diarrhoea in Indonesia: Effects of household and community factors

ABSTRACT

Diarrhoea is an important health issue in low- and middle-income countries, including Indonesia. We applied a multilevel regression analysis on the Indonesian Demographic and Health Survey to examine the effects of drinking water and sanitation facilities at the household and community level on diarrhoea prevalence among children under five (n = 33,339). The role of the circumstances was explored by studying interactions between the water and sanitation variables and other risk factors. Diarrhoea prevalence was reported by 4820 (14.4%) children, who on average were younger, poorer and were living in a poorer environment. At the household level, piped water was significantly associated with diarrhoea prevalence (OR = 0.797, 95% CI: 0.692–0.918), improved sanitation had no direct effect (OR = 0.992, 95% CI: 0.899–1.096) and water treatment was not related to diarrhoea incidence (OR = 1.106, 95% CI: 0.994–1.232). At the community level, improved water coverage had no direct effect (OR = 1.002, 95% CI: 0.950–1.057) but improved sanitation coverage was associated with lower diarrhoea prevalence (OR = 0.917, 95% CI: 0.843–0.998). Our interaction analysis showed that the protective effects of better sanitation at the community level were increased by better drinking water at the community level. This illustrates the importance of improving both drinking water and sanitation simultaneously.

KEYWORDS:

• Diarrhoea
• drinking water
• sanitation
• multilevel analysis
• Indonesia

Introduction

The latest UNICEF & WHO (2015) report on sanitation and drinking water worldwide indicates that over 663 million individuals still lack access to safe drinking water and 159 million persons rely on surface water for their water consumption. Within countries, the regional disparity in water access is substantial. About 79% of people depending on unimproved drinking water and 93% depending on surface water live in rural areas (UNICEF & WHO, 2015). In addition, 2.4 billion people (32%) worldwide lack access to improved sanitation facilities. In these situations, with poor water quality and high contamination risk, diseases such as diarrhoea become a major concern. Diarrhoea is still one of the most important killers of children under five (WHO & UNICEF, 2013a).

Indonesia closely reflects this global pattern. A considerable proportion (18%) of Indonesian households rely for their drinking water on surface water sources, such as springs, rivers, ponds and lakes, which are prone to contamination problems (Statistics Indonesia, 2014). Only 11% of Indonesian households have access to piped water inside their dwelling (Statistics Indonesia, 2014), and even then the quality is often below the minimum requirement for drinking water, with fluctuating debit and frequent interruptions (Surjadi, 2003). Sometimes the piped water is contaminated with faecal coliform and unsafe to be consumed without processing steps (Bakker, 2007). Almost all households boil their drinking water (Prihartono et al., 1994), but this is not always done effectively, as 55% of drinking water samples were found to be contaminated with faecal coliform (Vollaard et al., 2004).

In addition, Indonesia has the second highest number of people (54 million) in the world that practice open defaecation (UNICEF & WHO, 2015). This increases the risk of environmental pollution and water contamination even more. Given the poor quality of water and sanitation, it comes as no surprise that diarrhoea is still a major health concern in Indonesia, responsible for 31% of post-neonatal mortality and 25% of child mortality (UNICEF, 2012).

Whether a child suffers from diarrhoea is influenced by many factors, at the level of the household as well as at the level of the community in which the household is living. Ideally, an analysis of the determinants of diarrhoea should take all relevant factors at both levels into account (Corsi et al., 2011; Fewtrell et al., 2005). Children from a household with good quality water and sanitation are still at risk for diarrhoea if they live in a community with open defaecation, due to the contamination of soil and water sources (Andres, Briceño, Chase, & Echenique, 2014; Corsi et al., 2011). By focusing the analysis only on factors at the household level, an incomplete picture is obtained and interventions might appear to be less effective than they truly are (Corsi et al., 2011).

From a policy perspective, it is very important to know whether the effects of risk factors vary across circumstances. If it is known under which circumstances a certain risk factor or protective measure is important, policy-makers can tailor interventions towards the requirements of the specific situation. We can find these circumstances by including interaction terms between the risk factor or protective measure and variables indicating the circumstances in our analysis. However, until now no encompassing study on diarrhoeal disease in Indonesia has been published in which both direct and interaction effects of the major risk factors are studied simultaneously.

The current study aims to fill this gap in our knowledge by examining the impact on childhood diarrhoea of the quality of water and sanitation at household and community level in Indonesia. The model includes interactions between the risk factors and variables describing the specific situation. We perform a multilevel analysis on data from the 2007 and 2012 Demographic and Health Surveys (DHS) in order to answer the following research questions:

1. Does access to and treatment of water and sanitation at the household level influence diarrhoea prevalence?

2. Does water and sanitation coverage at the community level influence diarrhoea prevalence?

3. Under what circumstances are water access and sanitation important for preventing diarrhoea?

We expect that better quality of water and sanitation protect children against diarrhoea and that the relevant factors are found at both the household and community level. In addition, we expect that larger effects will be observed for children living in more deprived circumstances, such as in poor households and in communities with poor hygiene.

Conceptual framework

Our theoretical framework assumes children's health condition to be influenced by individual, household and community factors (Figure 1). At the household level, we focus on access to drinking water, on whether the water is treated, and on the quality of the sanitation facilities (arrow A in Figure 1). In Indonesia, many households still rely on unsafe drinking water, such as water from unprotected wells and rivers that are vulnerable to microbial contamination. This contamination takes place not only at the water source, but also during collection, transport, storage, and serving of the water, due to faecally contaminated hands, utensils and insects (Shaheed, Orgill, Montgomery, Jeuland, & Brown, 2014).

Figure 1. The impact of the household and community level of water and sanitation on childhood diarrhoea.

Piped water on the premises is expected to be less contaminated, as by the nature of its construction the piped water system protects against outside influences (WHO & UNICEF, 2013b). In addition, households connected to piped water can improve their health outcomes because more water is available for cleaning the house, thus facilitating a better hygienic situation (Fewtrell et al., 2005). Connection to piped water does not, however, always guarantee better water quality, as in less developed regions the water is often not continuously running. This means that households still have to store water in the home that is then vulnerable to (re)contamination (Fewtrell et al., 2005; Shaheed et al., 2014; Wright, Gundry, & Conroy, 2004).

When the quality of the available water supply is less than ideal, treating the water by boiling, chlorinating, filtering or other methods is an important behavioural strategy for reducing the risk of diarrhoea (Clasen, Schmidt, Rabie, Roberts, & Cairncross, 2007). Point-of-use water treatment improves the microbial safety of the water before consumption (Sodha et al., 2011) and reduces the risk of diarrhoeal diseases, specifically in developing countries (Fiebelkorn et al., 2012). However, the benefits of this treatment are not guaranteed, as the cleanliness of the treated water is often not maintained during storing and serving. It might for instance be touched while being put in or removed from the containers, which reduces the protective effects of treatment (Sodha et al., 2011; Wright et al., 2004).

Besides clean water, a good sanitation facility is protective against diarrhoea. Such a facility separates the human excreta from direct contact with humans and ensures a safe disposal of the faeces, thus reducing the risk of faecal contamination (Andres et al., 2014). However, as shown by a recent cluster-randomised trial in rural Odisha, India, proper sanitation facilities at the household level do not always improve health (Clasen et al., 2014). Exposure to faecal contamination in the community can wipe out the beneficial effects of good facilities at home.

Arrow B in Figure 1 highlights the importance of water and sanitation at the community level for children's health status. Health outcomes of children in households with good water and sanitation might be suboptimal when the hygiene level of the environment in which the household resides is low. Poor environmental hygiene may directly (e.g. by contact with contaminated water or human excreta in the open field) or indirectly (e.g. through contact with contaminated flies or other children) spread faecal contamination and water contamination over the community. In this way, members of households that already have good water and sanitation may be affected (Corsi et al., 2011; Gragnolati, 1999). Hence, important spillover effects between factors at the household and at the community level can be relevant.

Positive spillover effects may occur if households with access to clean water share their water with neighbours that have no such access (Alderman, Hentschel, & Sabates, 2003). This might facilitate better hygiene and health outcomes for children in the community (Corsi et al., 2011; Gragnolati, 1999). An even stronger spillover effect is expected with respect to the coverage of proper sanitation in the community (Alderman et al., 2003). Although an improved toilet increases the hygiene level of the owner, it cannot fully eliminate faecal contamination from the neighbourhood if other households lack such a facility (Andres et al., 2014; Clasen et al., 2014).

Indirect effects at the community level might also be important. Hughes and Dunleavy (2000, as cited in Alderman et al., 2003), for example, found that positive health effects of community access to water only materialised when combined with the community access to toilet. We therefore include in our model interaction terms between the water and sanitation variables at household and community level (arrows C in Figure 1).

Control factors

Our model contains a number of control factors that are known or expected to influence diarrhoea risk (arrows D in Figure 1). At the individual level, the child's age and sex are important. Approximately, 80% of the total diarrhoea incidence relates to children younger than two years of age (Walker et al., 2013). After this age, the rate of illness falls as the first infections induce a certain level of immunity, which protects against following incidences (Yu et al., 2015). Gender is also important, because of differences in immune system functioning between girls and boys, and because girls have been shown to have lower morbidity and mortality rates than boys (Muenchhoff & Goulder, 2014).

At the household level, we include mother's education and household wealth. An educated mother may have a higher level of awareness about hygiene (Alderman et al., 2003) and be better able to obtain clean water and to treat water effectively (Mangyo, 2008). Poorer households have fewer resources to fulfil their basic necessities, have poorer living conditions and have a lower health status, all factors that increase diarrhoea risk (Hatt & Waters, 2006).

Control factors at community level are availability of health facilities, level of regional development, urbanisation, adult education and the position of women in the area. Households in more developed or urban regions can generally benefit from better infrastructure, including more and better health facilities. Living under better circumstances in terms of infrastructure and health services may benefit the households in the area, including their children (Fotso & Kuate-Defo, 2005). Given that women are the major caretakers of young children, a stronger position of women in the household and local community might be favourable for children's health. We therefore also control for the decision-making power of women.

Methods

Data

This study utilised the Indonesia DHS from 2007 and 2012 (Statistics Indonesia & Macro International, 2008; Statistics Indonesia, NPaFPBB, MOH, & ICF International, 2013). These DHS surveys were designed to be representative at the national, urban, rural, as well as provincial level. Both data sets collected information on demographic, socioeconomic and health-related issues. To ensure the protection of human subjects, all DHS protocols are reviewed by an ethics review panel or institutional review board in the country where the survey is conducted (ICF International, 2012).

For the current study, the surveys were pooled into a single data set. This combined data set includes data on 33,399 children under age 5, from 28,573 mothers, living in 3069 sub-districts, within 922 districts, within 33 provinces. We excluded children with missing data on diarrhoea infection (N = 338) and children with missing response(s) on explanatory variable(s) (N = 1521 or an additional 4.4% of children).

Outcome variable

The outcome variable is a dummy variable indicating whether or not the child suffered (yes = 1, no = 0) from diarrhoea in the past two weeks. This question was asked of all mothers with living children under age five.

Independent variables

Independent variables were included at the household and sub-district level. A sub-district is the second lowest pubic administration before village level. On average, it covers an area of 273.6 km² and has a population of about 36,000. Variables at sub-district level were aggregated from household data. They were calculated as the proportion of individuals or the mean of the variable within each sub-district.

The main household level predictors were variables indicating whether there was piped water in the dwelling (yes, no), whether point-of-use water treatment was used (yes, no), and whether the household had improved sanitation (yes, no). Point-of-use water treatment indicated whether the household used any treatment method, such as boiling, bleaching, chlorinating, filtering or solar disinfection to the water before its consumption. Improved sanitation indicated a toilet facility that ensured the separation of human excreta from human contact, such as a toilet with septic tank, pit latrine or composting toilet (UNICEF & WHO, 2015).

Other control variables were the child's gender (girl, boy), age (0–4 years) and age-squared, mother's education (years of schooling completed), household wealth and dummy variables indicating whether the household lived in an urban (1) or rural (0) area and whether the household was interviewed in the 2007 (0) or 2012 (1) DHS survey. Household wealth was measured by an index constructed following Filmer and Pritchett (2001) by applying a principal component analysis (PCA) on the following variables: indicators of whether the household owned a radio, television, refrigerator, telephone, car, bicycle or motor bike, whether there was electricity in the dwelling and the quality of the floor material used for the dwelling. In line with Smits and Steendijk (2015), the outcome of the PCA analysis was translated into a continuous index ranging from 0 (having none of the assets and lowest floor quality) to 100 (having all assets and highest floor quality). We could not use the regular DHS wealth index available in the data set (Rutstein & Johnson, 2004), as it uses information on the quality of drinking water supply and toilet facilities, which we wanted to study separately in our analyses.

At the sub-district level (i.e. community level), we included seven contextual variables. Improved water coverage was measured by the proportion of households with improved drinking water, i.e. piped water on premises, public taps, tube wells, protected dug well, protected springs and rainwater collection in the community (UNICEF & WHO, 2015). Improved sanitation coverage, as a proxy for environmental hygiene, was measured by the proportion of households with an improved toilet facility in the community. Health facilities coverage was indicated in line with Monden and Smits (2009) as the proportion of mothers who gave birth in a proper health facility, such as a hospital, health centre, village health post or with help of a village midwife. Another health indicator at the community level was the proportion of children who received three polio vaccinations. The community level of economic development was indicated by the proportion of households owning a car. Two context variables – adults’ education and maternal decision power – were created to measure the availability of (health) knowledge and the strength of the position of women in the community. Adult education was measured by the average years of education completed by adults aged 15 and over in the community. Maternal decision power was measured by the proportion of mothers who reported that they could decide by themselves whether a child should be taken for medical treatment. Further description of variables used in the analysis can be found in the Appendix, Table A1.

Statistical analysis

We used a four level multilevel logistic regression model, with households (level 1) nested in sub-districts (level 2), nested in districts (level 3), and nested in provinces (level 4). Given that the average number of children per household was very small (1.2), children and households are considered as part of the same level (level 1). The model contains explanatory variables at levels 1 and 2. Levels 3 and 4 are included as random effects. The model was estimated with MLWin version V.2.29, using second-order PQL, the recommended estimation technique for multilevel logistic regression analysis (Goldstein, 2011). Both bivariate and multivariate multilevel models were estimated. Interaction analysis was used to study how the five main variables (piped water in the dwelling, point-of-use water treatment, improved toilet, improved water coverage and improved sanitation coverage) varied across circumstances. Given the explorative nature of the interaction analysis, we tested for all potential interactions between the main variables and the other variables in the model. To be able to focus on the most important interaction effects, only significant interactions were included in the final model. In this way, a parsimonious picture is obtained of the way in which the effects of the independent variables differ between children living under different circumstances. In the interaction analyses, centred versions of the variables were used. The main effects therefore can be interpreted as average effects. Statistical significance was evaluated at p < .05 and the coefficients are presented as odds ratios (OR).

Results

Table 1 presents descriptive of the data. The mothers of 4820 (14.4%) children reported that they suffered from diarrhoea in the preceding two weeks. Most households have no access to piped water in their premises (86.82%), treat the water (78.93%) and have an improved toilet (58.33%). Of the 33,399 children, 16,010 (47.94%) are girls, and 16,925 (50.68%) are from the 2007 DHS data set. The average age of the children is about two years. About 50% of the households in the community have access to improved water and 59% to improved sanitation.

Table 1. Descriptive characteristics of children aged under five in Indonesia, DHS, 2007 and 2012.

Variables	n/mean	%/(SD/range)
Total	33,399	100.00
Children having diarrhoea	4820	14.43
Children with piped water in the dwelling	4402	13.18
Children with point-of-use water treatment	26,361	78.93
Children with improved toilet	19,480	58.33
Gender is girl	16,010	47.94
Mean children age (SD/range)	1.98	(1.42/0–4)
Mean years of mother education (SD/range)	8.67	(3.85/0–15)
Mean household wealth (SD/range)	51.62	(25.94/0–100)
Children living in urban area	14,037	42.03
Children from DHS 2012	16,474	49.32
Context factors
Mean improved water coverage (SD/range)	0.50	(0.29/0–1)
Mean improved sanitation coverage (SD/range)	0.59	(0.30/0–1)
Mean health facilities coverage (SD/range)	0.48	(0.37/0–1)
Mean vaccination coverage (SD/range)	0.65	(0.24/0–1)
Mean economic development (SD/range)	0.07	(0.10/0–1)
Mean adults’ education (SD/range)	7.94	(2.25/0.26–16)
Mean mother's decision power (SD/range)	0.82	(0.18/0–1)

The results of the bivariate analysis are presented in Table 2. The figures in this table show how diarrhoea prevalence varies among households and communities with different characteristics. Most of the coefficients are significant and support the hypothesis that improved water and sanitation reduce the risk of childhood diarrhoea. Households with piped water or with an improved toilet facility have lower odds of diarrhoea. Interestingly, water treatment is associated with a higher risk of diarrhoea among children.

Table 2. Bivariate multilevel analysis of the diarrhoea prevalence of children aged under five, Indonesia, 2007 and 2012.

	OR (95% CI)
Piped water in the dwelling
No (ref)	1.000
Yes	0.748 [0.655–0.854]*
Point-of-use water treatment
No (ref)	1.000
Yes	1.110 [1.003–1.227]*
Improved toilet
No (ref)	1.000
Yes	0.826 [0.762–0.895]*
Gender
Boy (ref)	1.000
Girl	0.795 [0.740–0.853]*
Child's age	1.264 [1.159–1.378]*
Child's age square	0.883 [0.863–0.902]*
Mother's education	0.979 [0.968–0.989]*
Household wealth	0.844 [0.807–0.883]*
Living in an urban area
Rural (ref)	1.000
Urban	0.870 [0.785–0.966]*
Survey year
2007 (ref)	1.000
2012	1.076 [0.961–1.205]
Context factors
Improved water coverage	0.951 [0.904–1.001]
Improved sanitation coverage	0.830 [0.778–0.875]*
Health facilities coverage	0.816 [0.772–0.863]*
Vaccination coverage	0.865 [0.821–0.911]*
Economic development	0.865 [0.819–0.914]*
Adult education	0.860 [0.814–0.907]*
Mother's decision power	0.930 [0.884–0.979]*

Note: Ref, reference category.

*p < .05.

As expected, girls have a lower risk of having diarrhoea. The significance of age and its squared term suggests that the relationship between age and the odds of having diarrhoea is curvilinear with increasing prevalence in the first two years of life and decreasing prevalence afterwards. Mother's education, household wealth and living in an urban area all are associated with significantly lower odds of having diarrhoea. There is no significant difference in diarrhoea risk between the two survey years.

Except for improved water coverage, the bivariate model shows that all community characteristics are significantly associated with diarrhoea prevalence and that the direction of their effects is in line with our hypotheses. The likelihood that children have diarrhoea is lower when more households in the community have improved water and sanitation, and the community is characterised by better health facilities, a wealthier or a better-educated population, and women with more decision-making power.

Multivariate analysis

Results of the multivariate analysis are presented in Table 3. Model 1 includes only the main effects and model 2 includes both the main and interaction effects.

Table 3. Multilevel analysis of the diarrhoea prevalence of children aged under five, Indonesia, 2007 and 2012.

	Model 1 – no interaction	Model 2 – with interaction
	OR (95% CI)	OR (95% CI)
Intercept	0.134 [0.115–0.157]*	0.137 [0.117–0.160]*
Piped water in the dwelling	0.797 [0.692–0.918]*	0.763 [0.643–0.907]*
Point-of-use water treatment	1.106 [0.994–1.232]	1.089 [0.977–1.214]
Improved toilet	0.992 [0.899–1.096]	0.951 [0.860–1.052]
Girls	0.791 [0.736–0.850]*	0.793 [0.738–0.852]*
Child's age	1.265 [1.159–1.379]*	1.276 [1.170–1.392]*
Child's age square	0.882 [0.863–0.902]*	0.880 [0.861–0.900]*
Mother education	0.994 [0.981–1.006]	0.996 [0.983–1.008]
Household wealth	0.905 [0.853–0.959]*	0.890 [0.839–0.944]*
Living in urban area	1.177 [1.035–1.338]*	1.223 [1.075–1.392]*
Survey year	1.261 [1.119–1.421]*	1.271 [1.128–1.432]*
Context factors
Improved water coverage	1.002 [0.950–1.057]	1.013 [0.958–1.071]
Improved sanitation coverage	0.917 [0.843–0.998]*	0.898 [0.825–0.977]*
Health facilities coverage	0.870 [0.806–0.940]*	0.858 [0.795–0.926]*
Vaccination coverage	0.944 [0.891–1.001]	0.947 [0.893–1.005]
Car ownership	0.947 [0.886–1.012]	1.001 [0.933–1.074]
Adult education	1.074 [0.981–1.174]	1.057 [0.966–1.157]
Mother decision power	0.961 [0.913–1.012]	0.970 [0.921–1.022]
Interaction effects
Piped water in the dwelling × living in urban area		2.142 [1.490–3.078]*
Piped water in the dwelling × health facilities coverage		0.704 [0.588–0.842]*
Piped water in the dwelling × vaccination coverage		1.267 [1.063–1.512]*
Piped water in the dwelling × adult education		0.819 [0.685–0.978]*
Improved toilet × child age		0.906 [0.858–0.958]*
Improved toilet × mother decision power		1.090 [1.005–1.183]*
Improved toilet × improved water coverage		1.102 [1.005–1.207]*
Improved water coverage × car ownership		1.154 [1.080–1.233]*
Improved water coverage × improved sanitation coverage		0.914 [0.860–0.972]*
Improved sanitation coverage × household wealth		0.922 [0.879–0.968]*

Note: Ref, reference category.

*p < .05.

After controlling for other factors, the prevalence of diarrhoea is lower among children with piped water in the premises (OR = 0.80, 95% CI: 0.69–0.92) and, to our surprise, higher when the water is treated (OR = 1.11, 95% CI: 0.99–1.23), whereas no significant effect is found for an improved toilet in the household (OR = 0.99, 95% CI: 0.90–1.10). At the community level, the odds of having diarrhoea decrease as the coverage of improved sanitation in the community is higher (OR = 0.92, 95% CI: 0.84–1.00) while no significant effect of the coverage of improved water on diarrhoea prevalence (OR = 1.00 95% CI: 0.95–1.06) is found.

Children's age and age-squared are significantly related to diarrhoea risk, with a similar curvilinear shape as in the bivariate model. Also, the gender difference remains important, with lower prevalence among girls. At the household level, the effect of wealth remains significant, with less diarrhoea in wealthier households, but maternal education loses its significance. At the community level four variables – vaccination coverage, economic development, adult education and mother's decision power –lose their significance, while the other effects remain unchanged.

Interaction effects

To explore how the effects of the five core variables (piped water in the dwelling, water treatment, improved toilet, improved water coverage at community level and improved sanitation coverage at community level) vary across contexts, we computed interactions between these variables and the other factors in the model. These interactions were iteratively tested and the significant interaction effects were included in the model. This explorative procedure produced 10 significant interactions that were included in model 2 of Table 3.

The protective effect of piped water turns out to be stronger for children living in a community with good health facilities coverage and a more highly educated population (OR = 0.70, 95% CI: 0.59–0.84 and OR = 0.82, 95% CI: 0.69–0.98, respectively). This stresses the importance of clean water as a basic requirement for good facilities or available knowledge to be effective. The importance of piped water is further highlighted by the fact that in rural areas and areas that are badly covered by vaccination campaigns, the availability of piped water in the dwelling is associated with less diarrhoea. The difference between urban and rural areas is particularly strong (OR = 2.14, 95% CI: 1.49–3.08). Further analysis shows that in rural areas the OR is 0.55 (95% CI: 0.42–0.72). Consequently, in rural areas piped water is particularly important, while in urban areas piped water does not make a significant difference for diarrhoea risk (OR = 1.19, 95% CI: 0.95–1.48).

Improved sanitation at the household level reduces diarrhoea risk for older children (OR = 0.91, 95% CI: 0.86–0.96). This finding makes sense, given that babies and very young children do not yet use toilet facilities. The positive interactions of improved sanitation with maternal decision power (OR = 1.09, 95% CI: 1.01–1.18) and water coverage at the community level (OR = 1.10, 95% CI: 1.01–1.21) point towards the existence of compensatory effects, whereby better toilet facilities at the household level may compensate for lack of women's power and bad water coverage at the community level. In situations where women's power is at its lowest level, the effect of having a better toilet facility in the house is significantly negative (OR = 0.64, 95% CI: 0.44–0.95), hence associated with less diarrhoea. This is also true in communities with bad water coverage (OR = 0.81, 95% CI: 0.66–0.98).

Improved water coverage at the community level is most effective in reducing diarrhoea risk in less developed areas. It is also more effective if it goes together with better sanitation coverage at the community level and vice versa (OR = 0.91, 95% CI: 0.86–0.97), so that improvement of water and sanitation facilities in regions and villages should go hand in hand.

Finally, we see that the children of the wealthiest households profit most from improved sanitation at the community level. Although these children experience the most favourable circumstances at home, their risk of getting diarrhoea remains high as long as they live and play in a polluted community.

Discussion

The results of the current study indicate that having piped water in the dwellings reduces the odds of childhood diarrhoea by 24%. This percentage is similar to the 22% reduction in diarrhoea risk reported in a meta study (Fewtrell et al., 2005). However, we also find that piped water is particularly important in rural areas, where the circumstances are more difficult and hence improvement of water supply can make more of a difference. The fact that in urban areas the effect of piped water is weaker might have to do with the poor quality of the piped water system or with fluctuation debit. When the pressure in the pipes is low, polluted water from outside may enter and reduce the water quality (Shaheed et al., 2014). Another consequence of this fluctuating debit is that households need to store the water which again increases microbial (re)contamination that may increase diarrhoea risks (Wright et al., 2004).

The benefit from high coverage of health facilities complements the protective effect of piped water in the dwellings on childhood diarrhoea. Meanwhile, this protective effect is more effective in a community with low vaccination coverage. This indicates the complexity of diarrhoea disease, which can be transmitted not only through water but also through other routes such as food contamination (Agustina et al., 2013) that cannot be solved merely by the provision of safe water at the household level.

A surprising finding is that point-of-use water treatment is not significantly associated with diarrhoea prevalence and that there are no significant interactions with this factor. This result contrasts with research summarised in Fiebelkorn et al. (2012). It might indicate that in many Indonesian households water treatment is ineffective. This would be in line with previous studies (Sodha et al., 2011; Vollaard et al., 2004), which find that in Indonesia a significant proportion of the treated water is (re)contaminated by E. coli while being stored.

In contrast to previous studies (Fewtrell et al., 2005), the quality of the toilet facility at the household level has no direct effect, although we found some significant interaction effects. Having an improved toilet facility is associated with lower diarrhoea prevalence for older children, and in areas where mothers have less decision power, or improved water coverage is low. That older children (but still younger than five years) benefit from an improved toilet seems logical as these children are able to use toilet facilities independently.

The significance of the main effect of improved sanitation at the community level confirms the results presented in Buttenheim (2008). The effect of improved sanitation coverage is strengthened in communities with improved water coverage, which indicates that improvement of water and sanitation should go hand in hand to get the best results. This is consistent with the result reported by Hughes and Dunleavy (2000, as cited in Alderman et al., 2003) who found joint externality effects of community access to water and sanitation in rural India, while their direct effect was not detected.

In sum, the fact that we found 10 significant interaction terms illustrates the importance of studying the circumstances under which particular measures (piped water, better sanitation) are effective in reducing children's odds for getting diarrhoea. A limitation of this study is the use of a cross sectional design, which reduces the possibility to determine causal relationships. Lastly, this study focuses on Indonesia only, which could limit the applicability to other countries. However, the current study complements previous studies (Fuller, Clasen, Heijnen, & Eisenberg, 2014; Fuller, Westphal, Kenney, & Eisenberg, 2015), which document variation in the effect of water and sanitation on childhood diarrhoea between countries, by showing this variation between sub-districts in one country.

Conclusion

Diarrhoea is an important health issue in Indonesia as well as in many other developing countries. This study finds that piped water in the dwelling reduces diarrhoea risk for young children, whereas point-of-use water treatment and improved toilet appear to have no significant direct influence. This suggests that treatment is not done effectively and that it might be good that the government starts campaigns to make the public aware of this.

We hypothesised that the protective effect of water and sanitation facilities would be higher in poor situations. This appeared to be not always the case. We indeed found piped water at the household level to be particularly important in rural areas. However, the positive effect of better sanitation at the community level turned out to be stronger in communities with better water quality. This last finding illustrates the importance of improving drinking water and sanitation simultaneously and at the community level instead of some households only.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by a scholarship from the Directorate General of Higher Education (DIKTI), Ministry of Education and Culture, Republic of Indonesia.

Appendix

Table A1. Description of variables used in the analysis.

Variable	Description
Child level variable
Diarrhoea incidence	Equal to 1 if child suffered from diarrhoea in the past two weeks, 0 otherwise.
Gender	Equal to 1 if child was girl, 0 otherwise.
Child's age	Child's age in years (0–4).
Household level variable
Piped water in the dwelling	Equal to 1 if household had access to piped water in the dwelling, 0 otherwise.
Point-of-use water treatment	Equal to 1 if household treated the water before consumption by boiling, bleaching, chlorinating, filtering or solar disinfection, 0 otherwise.
Improved toilet	Equal to 1 if household had private toilet with septic tank, pit latrine or composting toilet, 0 otherwise.
Mother's education	Number of years of schooling completed by mother
Household wealth	Continuous variable of household's assets (1–100)
Living in an urban area	Equal to 1 if household lived in urban area, 0 otherwise.
Survey year	Equal to 1 if the year of DHS survey is 2012, 0 otherwise.
Community level variable
Improved water coverage	Proportion of households in the community with access to improved water sources (piped water on premises, public taps, tube wells, protected dug wells, protected springs and rainwater collections).
Improved sanitation coverage	Proportion of households in the community with access to improved sanitation facilities (toilet with septic tank, pit latrine, or composting toilet)
Health facilities coverage	Proportion of mothers in the community who gave birth in a proper health facility (hospital, health centre, village health post, or with help of a village midwife).
Vaccination coverage	Proportion of children in the community who received three polio vaccinations.
Economic development	Proportion of households in the community owning a car.
Adult education	Average years of education completed by adults aged 15 and over in the community.
Mother's decision power	Proportion of mothers in community who reported that they could decide by themselves whether a child should be taken for medical treatment.