2.1. Understanding barriers to toilet adoption in India

2.2. Mechanism linking improved sanitation and child health

This sub-section highlights the literature from biomedical sciences to understand the pathways through which water, sanitation, and hygiene interventions can lead to improved health. Three biological mechanisms link inadequate sanitation with undernutrition: diarrhea, intestinal infections, and environmental enteropathy [36]. Diarrhea is the result of infections from consuming food or water contaminated with feces, or even direct ingestion of fecal matter [37]. Soil-transmitted helminth infections are caused by soil contaminated with eggs present in human feces. These eggs are either ingested from nearby contaminated water sources or children’s soiled unwashed fingers after playing in the soil [38]. Repeated exposure to infections changes the structure of the gut, leading to impaired immunity and stunting without overtly manifesting into diarrhea [36]. However, the adverse consequences from the fecal-oral route can be prevented easily as the transmission occurs in the domestic and community environment of the child. It is in this context that hygiene and sanitation interventions play an important role in reducing the occurrence of diseases.

The F-diagram by Curtis, Cairncross, and Yonli, 2000 (see Fig 1) sketches all the potential pathways of transmission of fecal matter to the host [39]. It is also helpful to visualize the impact of interventions introduced at different stages. Primary barriers are employed to prevent the entry of feces into the environment, while secondary barriers stop the pathogens that have already entered the environment from infecting the host [39]. Hygiene promotion campaigns, in addition to promoting toilet construction, can help prevent environmental contamination and disease spread.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. From Fig 2 of Curtis, Cairncross, and Yonli, 2000.

https://doi.org/10.1371/journal.pone.0318198.g001

3.1. Research design and its contribution

Understanding the barriers to access and adoption of toilets in India has led us to a research design that incorporates all the limitations identified in previous studies:

1. Ensuring Access to High-Quality Individual Household Toilets (Removal of Credit Constraints): Past initiatives have struggled with inadequate coverage and substandard toilet construction, often leading to a high proportion of dysfunctional facilities. This research design emphasizes access to high-quality individual toilets for every household.
2. Fostering Toilet Ownership and Investment: Evidence suggests that when households are involved in the construction of their own toilets, they are more likely to adopt and use them consistently. Our approach aims to provide financial and in-kind contributions that promote a sense of ownership, thereby increasing the likelihood of sustained toilet use.
3. Implementing Universal, Rigorous Behavioral Change Communication (BCC) Campaigns (Eliminating Information Constraint): Previous campaigns have shown that behavioral change initiatives alone may not be sufficient to eliminate OD. Our research design will integrate a robust BCC strategy that not only engages all the members in the communities but also promotes a sustained cultural shift towards improved sanitation practices.

In partnership with the local non-governmental organization, Grameen Development Services (GDS), we implemented a cluster-randomized control trial (CRCT) in the Maharajganj district of eastern Uttar Pradesh, India, involving 15 villages and 866 households. This district was chosen as our 15 study villages served as part of Tata-Cornell Institute (TCI), Cornell University’s Technical Assistance and Research in Indian Nutrition and Agriculture (TARINA) (S. Gupta et al., 2019). Maharjganj district was also well suited to our study as according to NFHS-4 (2015-16) this district had only 17% toilet coverage in the rural areas and low exposure to any sanitation program. Further, the district head agreed to not undertake any sanitation intervention in the control villages until the completion of the experiment.

The randomization of this CRCT study design was done at the village level with an equal allocation of villages to the two treatment arms and one control arm. Five villages randomized to the cluster A arm received the community-level BCC campaign that uses the tenets of community-led total sanitation (CLTS), along with the construction of subsidized individual household toilets for every household that chooses to avail the subsidy. The five villages randomized to cluster B group were provided with the same choice of building toilets as was given to cluster A but without any BCC intervention. The five villages randomly assigned to the control group did not receive any intervention but participated in the survey rounds. All 866 households across the 15 villages participated in the study, representing a comprehensive census of the entire population. The study protocol was approved by the Institutional Review Board at Cornell University and informed consent from the respondents as well as the primary caretaker of the child was recorded electronically (Protocol number: 1609006597). The treatment arms are described in detail below.

Control group.

The control group comprised households that received no interventions. They participated solely in survey rounds, providing a baseline against which the effects of the interventions in Clusters A and B could be measured.

Both treatments of BCC campaigns and building of subsidized individual household toilets were assigned at the village level and given to all households. This is a common practice in the sanitation-nutrition literature as OD emanates negative externalities. That is, as long as some fraction of members in the community are practicing OD, their neighbors, even if using toilets, are still exposed to fecal pathogens and prone to infection [41]. Hence, to ensure that the elimination of OD happens at the community level, the interventions are targeted towards all the households in the village.

See Table 1 for a summarised overview of the interventions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Details of intervention by groups.

https://doi.org/10.1371/journal.pone.0318198.t001

Our research design is unique and contributes to the literature in several ways. First, our study recognizes the externalities imposed by OD as even the members of the community who do not practice OD still live in an environment exposed to fecal pathogens. Therefore, our program provides the option to build toilets for every household in the 10 treatment villages and by providing the necessary technical support to build good quality toilets (removal of credit constraints). Second, this design mandates that households bear the rest of the 25% of the cost giving them a stake in ownership and making them more likely to use the toilet [8,14,42,43]. Third, our research design acknowledges the widespread preference for open defecation and consequent complementarities between sanitation campaigns and toilet adoption. Hence, half of the treatment villages receive community-wide behavior change intervention about the issues of OD before the option of construction is presented to them (elimination of information constraint). Therefore, our treatment in cluster A imparts both universal access to toilets and information. Our study recognizes that ensuring high sanitation coverage and use as well as the provision of information to all the participants is the fundamental foundation of an effective sanitation strategy.

3.2. Experiment timeline

The female head of the household was our primary respondent in the survey. The study was conducted between 2016 and 2019 and its timeline is depicted in Fig 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Timeline of the intervention and data collection exercise.

https://doi.org/10.1371/journal.pone.0318198.g002

Prior to the intervention, we administered a baseline survey for all 866 households in the 15 villages in June-July 2017. It included modules on socioeconomic characteristics of the household, hygiene practices, defecation behavior, and preferences, among other questions. In addition, we also collected information on children’s nutrition outcomes as captured through anthropometric measurements for all 278 children below five years of age in the 15 villages. The BCC implementation exercise was completed by the end of July 2017 and toilet construction was finalized by January 31, 2018.

From February 2018 to December 2018, we conducted monthly surveys with households to track toilet adoption trends. These short follow-up surveys involved interviewing both the household heads and the primary female respondents about their toilet use. Additionally, enumerators verified toilet usage during these visits. This approach allowed us to capture variations in toilet adoption across different interventions, genders, and seasonal patterns, providing a comprehensive understanding of the trends over time. We did not seek information on other modules from the baseline survey in these monthly surveys.

The final endline survey was carried out in December 2018 and January 2019, 18 months after the behavior change dissemination and construction of temporary toilets, and around 11-12 months after the completion of toilet construction. The modules were similar to the baseline survey.

Of the 866 households covered in the baseline survey, 846 (97.6 percent) participated in the endline survey. Of the 846 households, 339 (40%) belonged to Cluster A, 270 (32%) belonged to Cluster B, and 237 (28%) in the control arm. Of the 278 children who were surveyed in the baseline, only 213 children who were less than five years of age in the baseline and endline were included for analysis. A panel dataset of the same set of 846 households in the baseline and endline and a panel of 213 children in the baseline and endline are used for the empirical analysis.

4.1. Descriptive statistics and balancing test

Table 2 presents the summary statistics of the sanitation profile of the respondents and the households as well as child characteristics from the baseline survey. More than 95% of the respondents were practicing OD and only 5% of the households had a toilet in the baseline. Next, even though around 40% of the households had soap near the handwashing station, only around 10% of the respondents reported washing their hands with soap after defecation. In terms of nutritional outcomes, around one-third of the children below five years of age in the program villages were underweight with a mean weight-for-age (WAZ) score being −1.45. 27% of the children reported having suffered from at least one diarrheal episode in the past 30 days.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Baseline summary statistics on the sanitation profile of the households and respondents.

https://doi.org/10.1371/journal.pone.0318198.t002

Table 7 in Appendix 2 in S1 File reports the baseline averages for a host of household, respondent, and children characteristics across the treatment arms. Overall, the sample is balanced as treatment and control villages are quite similar.

4.2. Empirical strategy

The random assignment of villages to different treatment arms provides a straightforward causal interpretation of each intervention. We employ a difference-in-difference strategy of the following form where . is the outcome for individual i, residing in household h, village v at time t:

Villages assigned to both BCC and toilet subsidy (cluster A) are captured through the dummy variable while indicates only toilet subsidy villages (cluster B). The coefficients and are the causal estimates of the intervention in both these clusters as compared to the control group. The key identification assumption here is that besides the two types of treatment, any time-varying factor will impact the relevant outcomes in the treatment and control groups in the same manner. Since d were randomly assigned, we expect and . Under the parallel trends assution, we can test the significance of the treatment through the null hypothesis that and .

We control for the vector of individual, household, and village-specific characteristics, . In trying to estimate the impact of our intervention on child nutrition, we control for the mother’s body mass index and education. Mother’s literacy level accounts for her agency and awareness of best nutritional practices. At the household level, we control for the household head’s age, education, religion, size of the landholding, and an index for durable assets. We also include gender dummies along with the child and village-level fixed effects ( and , respectively). Since the treatment was assigned at the village level, we cluster the standard errors by the village. We use a Wild-cluster bootstrapping procedure to bootstrap the standard errors on our coefficients of interest for improved inference with fewer clusters [44].

4.3. Outcomes

The study focuses on three primary outcomes (1) access to toilets, (2) open defecation, (3) hygiene practices, and their subsequent impact on one secondary outcome (4) child health outcomes.

1. To measure access to toilets in a village, we present results for the percentage of households in the village having access to a toilet.
2. Open defecation is measured through the percentage of respondents practicing OD or not using toilets. To ascertain this, we asked the respondents where they went for defecation in the previous 24 hours.
3. Our BCC intervention also influenced the hygiene behavior outcomes. Hygiene behavior was measured through the following outcomes: handwashing with soap (self-reported), presence of soap at the handwashing station, proper stool disposal, and frequency of using toilets (or going for OD).
   1. Responses for handwashing with soap were classified as a dummy: 1 for yes, and 0 for no.
   2. Similarly, the presence of soap at the handwashing station was captured through an indicator.
   3. Proper stool disposal (i.e., either flushing the child’s feces in the toilet or burying it in the ground) was captured through a binary indicator 1 for proper stool disposal and 0 for improper stool disposal.
   4. Finally, the frequency of going for urination or defecation was also included in the results as it baseline survey suggested that it varied from people practicing OD and the ones using toilets. People usually go to the field to urinate and/or defecate twice a day (either before sunrise or after sunset) in rural India. Access to toilets grants people the freedom to use them for urination or defecation at any time during the day. Hence, we constructed an indicator variable that captures if people went for urination or defecation at most twice a day (variable takes the value 0), and more than twice in one day (variable takes the value 1).
4. We measure the child’s nutritional outcomes through changes in weight.

The weight of a child is measured using a weight-for-age z score (WAZ score), which compares the child’s weight to the weight of healthy children of the same age and sex in a reference population, in terms of standard deviations. It is defined as

The WAZ score is useful to identify whether the child is underweight, normal weight, or overweight. A WAZ score of zero indicates that the child's weight is the same as the median weight of the reference group. A negative WAZ score means that the child's weight is lower than the median weight of the reference population. If a child's WAZ score is two standard deviations below that of the reference group, they are classified as underweight, which reflects a nutritional failure. To ensure accuracy, we follow the WHO's 2006 international reference population and exclude children whose WAZ scores are below minus six or above five standard deviations [45]. We further check for potential pathways of the increase in WAZ scores by using diarrhea prevalence for analysis.

5.1. Impact of treatments on ownership of individual household toilets and open defecation

Our findings report an increase in the percentage of households that built individual household toilets (from 2% to 98%) (or an increase from 13 to 332 toilets) (Fig 3a) and consequently led to a decline in respondents practicing OD in cluster A (from 98% to 4%) (or a decrease from 333 to 13 female respondents) (Fig 3b). There was also a near-universal decline in OD for the male respondents of the households at the endline in cluster A (see Fig 3, Appendix 3 in S1 File). All female respondents from cluster A villages who resided in households that had a toilet reported that all their family members used the toilets as well, in the endline.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. (a) and (b): Impact of treatments on access to toilets and open defecation.

https://doi.org/10.1371/journal.pone.0318198.g003

In cluster B, the percentage of households with access to toilets improved by 81 percentage points (from 6% to 87%), but the percentage of respondents practicing OD fell only by 43 percentage points (from 95% to 52%).

In the control group, even though the individual household toilet access increased by 4.7%, the ODF rates remained about the same. This suggests that the increase in toilet adoption is not commensurate with the increase in toilet access, implying the need for hygiene promotion.

5.2. Impact of treatments on hygiene practices

Our BCC intervention is also hypothesized to have an impact on hygiene behavior outcomes. The DID estimates suggest there is a significant impact on hygiene practices. We find a 93% increase in self-reported handwashing with soap in cluster A after defecation relative to the control group (see Columns 1 and 2 of Table 3). The results for cluster B although significant are small in magnitude (relative to the control group).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Treatment impacts on hygiene behavior.

https://doi.org/10.1371/journal.pone.0318198.t003

Enumerator observation also corroborates this finding (although not to the same magnitude), as only 62% of households in cluster A had a soap present at their handwashing stations (see Columns 3 and 4 of Table 4). The difference could partly be explained as many respondents use washing detergents to wash hands after defecation. For our self-reported measure, we only counted handwashing with soap, and not detergent.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Impact of treatments on child weight-for-age z-scores.

https://doi.org/10.1371/journal.pone.0318198.t004

We also find a significant improvement in stool disposal as 83% of respondents in cluster A practice proper disposal techniques (i.e., either flushing in the toilet or burying it in the ground), relative to the control group. The results for cluster B are small and insignificant using Wild cluster bootstrapping (see Columns 5 and 6 of Table 3).

The last two columns of Table 3 estimate the results from an indicator variable that captures if the respondents practice urination or OD at most twice a day (variable takes the value 0), or more than twice in one day (variable takes the value 1). The estimates suggest that on average, 91% more respondents in Cluster A, and 43% more respondents in Cluster B practiced urination and OD more than twice in one day (relative to the control group).

5.3. Impact on child weight

OD can lead to the contamination of water and food supplies with fecal matter [46]. To reduce OD, it is necessary to have access to and use improved sanitation facilities that prevent human waste from re-entering the environment. Studies have demonstrated that interventions aimed at preventing human waste from entering the environment can decrease the incidence of diarrheal diseases and other infectious diseases as well as child underweight or low weight-for-age z-scores [3,47,48].

We test for similar results in our study, by estimating the causal impact of the treatments on child WAZ scores using regression estimates. DID regression results in Table 4 exhibit the impact of the two treatments relative to the control group.

Column (1) is a simple difference-in-difference estimate. The coefficient of 0.69 suggests that as the community moves from universal OD to no OD, the WAZ score for children below significantly increases by 0.69 standard deviations for cluster A while remaining insignificant for cluster B. Both estimates are relative to the control group. Adding controls for the mother’s BMI and education leads to a minor decrease in coefficient from 0.69 to 0.68 in column (2). As we progress from column (2) to (3) we add household controls, and the coefficients remain unchanged. Adding village fixed effects in column (4) also makes no change to the coefficient, which is unsurprising as the villages received treatments randomly. All specifications include female child dummies and child-fixed effects. The level of significance for these coefficients remains the same for all the columns. The coefficients for cluster B remain insignificant for all the specifications. Therefore, on average, the WAZ scores increased by about 0.68-0.69 standard deviations for children below five in cluster A relative to the control group. Estimates for impact in cluster B are smaller and insignificant. The WAZ coefficient is not just statistically significant for cluster A but remains robust as we progressively include more controls and fixed effects.

Our estimated coefficient size of 0.68-0.69 increase in WAZ scores in cluster A echoes well with the magnitude of other studies that analyze the impact of the elimination of OD on child anthropometry outcomes. In Cambodia, the elimination of OD at the community level is associated with a 0.56 increase in WAZ scores for children below five [49]. Dickinson et al. 2015 report that India’s national sanitation campaign, called the Total Sanitation Campaign, led to an increase in the ownership of toilets from 7% to 35% which led to a 0.26-0.31 increase in children’s WAZ scores, after one year of implementation of the program. The lower magnitude of this coefficient could be due to lower increase in toilet ownership and also that their follow-up period was around six months in treatment villages, as opposed to almost one year in ours. Another study conducted in rural Mali with children below five years found that children who got the CLTS intervention had a 0.18 higher height-for-age z-score (HAZ) score than the control group, 1.5 years after the intervention [19]. Similarly, a study conducted in India found that the impact of the Total Sanitation Campaign (India’s former sanitation campaign) led to an increase of 0.30 HAZ scores, 1.5 years after the intervention [7].