Introduction
The past decade has witnessed a renewed global interest in the health impacts of improved water, sanitation, and hygiene (WASH). The Millennium Development Goal (MDG) era saw solid progress in WASH indicators, with almost 2 billion people gaining access to improved water and/or sanitation. However, some 700 million still lack access to improved water; in addition, approximately 2.5 billion people do not use an improved sanitation facility, and of these, 1 billion people still practice open defecation (WHO and UNICEF
2014).
The persistence of these problems is a significant public health concern. Human feces are an important reservoir for a range of pathogenic bacteria as well as soil-transmitted helminths (STHs) that can cause diarrhea, environmental enteric disorder (EED), trachoma, and other morbidities prevalent in young children (Mara et al.
2010). Diarrhea and EED are also thought to be important determinants of malnutrition in young children (Checkley et al.
2008; Humphrey
2009). Moreover, many of these morbidities, in combination with poor nutrition, often prove fatal if not properly treated, suggesting that poor WASH conditions could be a major underlying risk factor for child mortality (Mara et al.
2010).
Yet, despite several plausible biological pathways, the empirical evidence linking WASH conditions to child health outcomes is limited and, for some health outcomes, inconsistent. Evidence from cluster randomized control trials (RCTs) and case-control studies suggests reasonably strong and consistent impacts of WASH interventions on diarrhea incidence (Fewtrell et al.
2005; Freeman et al.
2017; Wolf et al.
2014) and STH infections (Freeman et al.
2017; Strunz et al.
2014; Ziegelbauer et al.
2012). However, RCT estimates of WASH impacts on child stunting and wasting are often statistically insignificant (Dangour et al.
2013; Freeman et al.
2017). In contrast, observational research has typically found very strong associations with child health and nutrition outcomes. A range of historical studies have linked reductions in child mortality to WASH improvements in the nineteenth and early twentieth centuries (Cutler and Miller
2005; Woods et al.
1989). For 71 contemporary developing countries, a pooled multivariate regression analysis of Demographic Health Survey (DHS) data found that household water and sanitation facilities were strongly associated with lower risk of child mortality, diarrhea, and stunting (Fink et al.
2011). Various papers by Spears and colleagues also used DHS data to link child mortality, stunting, and anemia to toilet use within the broader community, on the premise that open defecation has negative interhousehold externalities on child health (Coffey et al.
2016; Geruso and Spears
2018; Spears
2013a). This research also uncovers evidence that open defecation may have more harmful effects in densely populated regions, such as South Asia (Hathi et al.
2017; Spears
2013a).
These different literatures therefore have tended to find reasonably strong evidence of WASH impacts on diarrhea, but impacts on child nutrition and mortality outcomes remain uncertain. In practice, both the experimental and observational literatures have important methodological limitations. Several commentaries raise concerns about the quality of the RCT evidence, highlighting issues such as the low adoption of WASH interventions and the short duration of exposure to WASH treatments (Headey
2016; Huda et al.
2012; Schmidt
2014). Still, observational studies also have inherent limitations. Most have used repeated cross-sections in which WASH exposure is not clearly linked to any specific intervention and is therefore likely to be strongly correlated with a range of confounding factors, including parental knowledge and preferences, cultural norms, local economic development, historical infrastructural investments, governance quality, and environmental factors, such as population density (Coffey et al.
2017; Davis
2004; Ndikumana and Pickbourn
2017). Adequately controlling for these interhousehold and intercommunity differences with cross-sectional survey data is likely to be extremely difficult, if not impossible. As a result, it is difficult to argue that these studies convincingly inform the more policy-relevant question that experimental studies pose: do
changes in WASH exposure lead to
changes in child health outcomes?
In this study, we use a subnational panel data set constructed from aggregated DHS to address this important policy question. Although the DHS are not a panel of children or households, they are a panel of subnational regions, the smallest geographical unit at which the DHS are representatively sampled. Moreover, DHS data on child health, sanitation, and other determinants of child health have been collected within countries in successive DHS waves over relatively long periods. These two features allow us to construct a rich subnational panel covering 442 subnational regions in 59 countries with multiple DHS rounds, resulting in approximately 1,500 observations for mortality, diarrhea prevalence, and fever prevalence, and 1,176 observations for stunting and wasting. This data structure has several key advantages.
First and foremost, it permits controls for panel fixed effects, thereby netting out the important time-invariant confounding factors listed earlier. Thus, we estimate difference-in-difference (DID) regressions that control for any non-time-varying subnational characteristics, regardless of whether they are observable in the data.
Second, subnational data exploit the growing importance of decentralized governance in developing countries. The importance of state-level changes in WASH in India and Nepal has been well documented (Coffey et al.
2016,
2017; Spears
2013b), but there are many other subnational WASH success stories. In Ethiopia, for example, the Southern Nations, Nationalities, and Peoples’ (SNNP) regional government implemented an exceptionally rapid expansion of community-led total sanitation over 2003–2005 prior to a national scale-up in 2006 (World Bank
2007).
Third, although changes in WASH access are not random in these data, DID regressions restrict endogeneity concerns to time-varying confounding factors, which we may be better able to adequately control for by including time-varying indicators from the DHS and other sources. Moreover, panel data permit us to assess to some extent—by exploring associations between the WASH variables and other likely determinants of child health and through parallel trends exercises—how likely it is that two of the likely sources of potential bias are driving the results.
Finally, in addition to addressing issues of internal validity, the geographical spread of DHS data allows us to speak to important issues of external validity, particularly whether the health benefits of expanded WASH access vary with population density (Hathi et al.
2017) or child age (Alderman and Headey
2018).
Our results suggest that changes in subnational sanitation coverage predict sizable improvements in child morbidity and mortality. A 1 percentage point increase in sanitation coverage is associated with a decrease in under-5 child mortality of between 0.34 and 0.38 per 1,000 births and a decrease in the prevalence of diarrhea during the two weeks preceding the survey of between 0.056 and 0.12 percentage points. In contrast, we find no statistically significant association between sanitation coverage and stunting or wasting, and the association with the prevalence of fever is highly sensitive to the specification used. Combining our estimates with the observed increase in global sanitation coverage between 1990–2015 indicates that changes in sanitation coverage can potentially explain 8.2 % of the total observed decline in under-5 mortality over the same period. We find little evidence that increases in access to any improved water source—according to the official definition—are statistically significantly associated with health and nutrition improvements. However, water piped into the home predicts significant reductions in child stunting, suggesting that the official definition of “improved water” may need to be revisited.
Discussion
WASH investments are widely viewed as an integral component of improving child health outcomes in developing countries. However, experimental evaluations of WASH interventions have not always uncovered strong evidence of impacts, particularly on child nutrition outcomes, and are also potentially subject to methodological limitations related to short timeframes, poor compliance, and limited external validity. These evaluations have not been statistically powered to assess precise mortality impacts. Instead, many researchers have resorted to observational analyses that exploit cross-sectional variation in water and sanitation access. Although such studies have generated useful suggestive evidence, cross-sectional estimates may be significantly biased by omitted time-invariant factors, offer few rigorous means of gauging that bias, and do not directly address the question of whether historical changes in WASH coverage typically lead to improvements in health outcomes.
In this study, we pursue a DID analysis to address some of the limitations in both the experimental and observational literatures. The subnational panel of DHS data used herein allows us to explore longer-term changes in WASH access in a broad swathe of countries, purge regressions of important time-invariant sources of bias, conduct a range of extensions and robustness tests, and conduct several falsification exercises.
At the same time, the data and methods used in this article are subject to limitations. Although the results pass most falsification checks, we cannot definitively rule out biases from time-varying omitted variables, which would caution against drawing overly strong causal inferences from these results. Our estimates are also somewhat imprecise and are therefore subject to uncertainty in a quantitative sense. We discuss and explore potential measurement issues with the dependent variables, but another source of imprecision is measurement error in the DHS WASH indicators. Sanitation indicators in the DHS are not ideal because toilet ownership does not always equate to toilet use or to appropriate disposal of children’s stools, although the fact that we find significant and relatively large coefficients on sanitation for two of the outcomes might suggest that attenuation bias is not an overwhelming problem for sanitation. Perhaps of greater concern is that improved water infrastructure could be a poor proxy for latent water quality in a microbial sense. For example, piped water systems that lack regular and consistent water flow may become breeding grounds for pathogenic bacteria (Klasen et al.
2012). Hence, there is likely to be important unobservable heterogeneity in the quality of piped water across countries. Still, the statistically significant association between water piped into the home and stunting—and the insignificant coefficients on improved water not piped into the home—suggests that the costs associated with collecting water outside the home may have especially harmful impacts on child welfare even with heterogeneity in water quality (Gross et al.
2018).
Another limitation is that our WASH indicators solely focus on hardware measures. Improving hygiene, however, is also likely to require significant behavioral changes that are not well recorded in the DHS and similar surveys. Formal education and adult literacy programs have been shown to be associated with both health knowledge and child health more broadly (Blunch
2013,
2017; Glewwe
1999; Kovsted et al.
2003), and it may be that this kind of soft knowledge complements the availability of improved WASH hardware.
Bearing these caveats in mind, many of our results are quite consistent with the experimental WASH literature. The importance of sanitation for reducing the prevalence of diarrhea accords closely to findings from both the experimental literature (Fewtrell et al.
2005; Freeman et al.
2017; Wolf et al.
2014) and the observational literature (Fink et al.
2011). Also consistent with much of the experimental literature is the lack of any statistically significant association between changes in sanitation and changes in child stunting and wasting (Dangour et al.
2013; Freeman et al.
2017).
16 There are plausible biological explanations for a relatively weak relationship between sanitation and stunting. Although some cohort studies have found that diarrhea episodes may contribute to stunting (Checkley et al.
2008), others have found that significant catch-up growth occurs after diarrhea episodes, thereby limiting long-run impacts on linear growth (Richard et al.
2014). Another recent line of research has speculated that animal feces may be an important contributor to EED and stunting (Headey and Hirvonen
2016; Headey et al.
2017; Mbuya and Humphrey
2016), an exposure unlikely to be influenced by conventional WASH hardware.
Despite disappointing evidence regarding sanitation’s impacts on child nutrition, we find relatively strong associations with child morbidity and mortality. We estimate that sanitation improvements have accounted for just under 10 % of the decline in child mortality from 1990 to 2015. This is a significant contribution, although because approximately 1 billion people still practice open defecation, further investments in sanitation are still very much needed.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.