The study was conducted in Somaliland, a self-declared state in the Horn of Africa with an estimated population of 5.7 million. Administratively, the country is divided into six regions: Awdal, Maroodijeex, Sahil, Togdheer, Sool, and Sanaag. These regions vary widely in socioeconomic conditions, geography, and access to health services. Undernutrition remains a major public health challenge across Somaliland, with persistently high rates of stunting, wasting, and underweight. Given the widespread reliance on unimproved water sources, poor sanitation facilities, and inadequate hygiene practices, Somaliland provides a critical context for examining the association between water, sanitation, and hygiene (WASH) indicators and child undernutrition (Figure 1).

The primary data for this study were extracted from the 2020 Somaliland Demographic and Health Survey (SDHS). The SDHS is a nationally representative household survey implemented by the Central Statistics Department (CSD) of the Ministry of Planning and National Development, in collaboration with the Ministry of Health Development. To ensure high data quality and international comparability, the survey design and data collection procedures strictly followed the global USAID DHS Program protocols. Data were collected using Computer-Assisted Personal Interviewing (CAPI) on tablet computers, which included built-in logic and consistency checks. The survey collected detailed information on demographic characteristics, maternal and child health, nutrition, and household environmental factors. Standardized DHS questionnaires were administered, and anthropometric measurements of children under 5 years of age were taken using calibrated equipment and procedures consistent with World Health Organization (WHO) recommendations. The SDHS dataset provides an opportunity to assess child nutritional outcomes in relation to household and community-level WASH conditions.

The SDHS 2020 applied a two-stage stratified cluster sampling design. In the first stage, enumeration areas (EAs) were selected using probability proportional to size from the national census frame. In the second stage, households were systematically chosen from each EA. All eligible women aged 15–49 years residing in the selected households were interviewed, and anthropometric data were collected for their children under 5 years of age. For this study, children aged 0–59 months with complete anthropometric data and available household WASH information were included in the analysis, resulting in a weighted sample of 1,195.

Data cleaning, recoding, and weighting were performed before analysis, and all statistical procedures were conducted using Stata version 17.0. Standard DHS weights were applied to the data to ensure the representativeness of the findings at the national and regional levels. Descriptive statistics were calculated to summarize sample characteristics and the prevalence of stunting and wasting. Logistic regression was used to assess associations between independent variables and child nutritional outcomes.

Given the hierarchical nature of the data, multilevel logistic regression models were employed to account for clustering of children within households and communities. Five models were fitted for each outcome variable. Model 0 was an empty model with no predictors, used to assess baseline variance and intra-class correlation coefficients (ICC). Model 1 combined individual-level characteristics with WASH indicators. Model 2 introduced community-level variables such as place and region of residence. Finally, Model 3 was the full model, which included WASH, individual, household, and community-level covariates. Results were reported as adjusted odds ratios (AOR) with 95% confidence intervals (CI). Model fit was evaluated using the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and log-likelihood statistics. Statistical significance was determined at p < 0.05.

This table, Multilevel Logistics Regression Analysis of wasting presents the Adjusted Odds Ratios (AORs) and 95% Confidence Intervals (CIs) from a series of logistic regression models (Model 1 to Model 4) examining the association between various WASH factors and wasting, after controlling for confounders. In Model 1, individual and household-level WASH and socio-demographic variables were introduced. Unimproved latrine facilities were significantly associated with child wasting, with children in households using unimproved sanitation having 30% higher odds of wasting compared with those using improved sanitation (AOR = 1.30; 95% CI: 1.00–1.60; p = 0.036). Open defecation showed increased odds of wasting; however, the association was not statistically significant. Use of improved drinking water sources was associated with lower odds of wasting, but this relationship did not reach statistical significance. Similarly, time to fetch water, whether less than 30 min or more than 30 min compared with on-premises access, showed no significant association with wasting. Regarding socio-demographic factors, female children had significantly higher odds of wasting compared with males (AOR = 1.20; 95% CI: 1.00–1.40; p = 0.017). Children living in nomadic households also exhibited higher odds of wasting (AOR = 1.50; 95% CI: 0.80–2.70; p = 0.023). Household head age and wealth quintile were not significantly associated with wasting in this model.

Model 2 examined community-level characteristics while excluding individual-level variables. Regional variations in wasting were observed, but none of the regions showed statistically significant differences compared with the reference region. Place of residence (rural or nomadic), community poverty level, community water coverage, and community sanitation coverage were also not significantly associated with child wasting. Although some point estimates suggested higher risk in nomadic and high-poverty communities, the wide confidence intervals indicate substantial uncertainty. The full model (Model 3) simultaneously adjusted for both individual- and community-level variables. The association between unimproved sanitation and wasting remained statistically significant, confirming its independent effect (AOR = 1.30; 95% CI: 1.02–1.68; p = 0.031). In contrast, open defecation was not significantly associated with wasting after adjustment, suggesting that its effect may be mediated through other household or community conditions. The association between female sex and wasting remained significant in the full model (AOR = 1.19; 95% CI: 1.00–1.30; p = 0.022). The previously observed elevated odds among children from nomadic households were attenuated and no longer statistically significant after controlling for community-level factors. Drinking water source, time to water source, household wealth, household head age, region, and community poverty all remained non-significant predictors of wasting in the fully adjusted model.

Across all models, unimproved sanitation emerged as the most consistent and robust WASH-related predictor of child wasting, maintaining statistical significance even after accounting for community-level influences. Female children were also at significantly higher risk of wasting, highlighting important sex-based disparities in acute undernutrition. While other WASH components such as water source and access time showed protective trends, their associations were not statistically significant after multilevel adjustment. Community-level factors explained part of the between-community variation but did not independently predict wasting. These findings underscore the critical role of household sanitation improvements in reducing child wasting in Somaliland, while also emphasizing the need for targeted interventions addressing gender-related vulnerabilities (Table 2).

Based on Table 3, Model 0 (empty model) shows significant community-level variation in wasting, with an ICC of 7.6%, indicating that a meaningful proportion of variability is attributable to differences between clusters. Model 1, which includes individual-level factors, slightly reduces the ICC and variance, suggesting that child- and household-level characteristics explain part of this clustering, although overall model fit does not substantially improve as reflected by a higher AIC and BIC. Model 2, which incorporates community-level factors, demonstrates the best model fit, as it has the lowest AIC (4264.643) and BIC (4327.952), along with a reduced ICC (6.7%) and lower random-effect variance. This indicates that community characteristics contribute importantly to explaining variation in child wasting and do so more efficiently than individual factors alone. Model 3 (full model) achieves the highest log likelihood, but its higher AIC and BIC indicate overfitting without proportional improvement in explanatory power. Overall, Model 2 is the best and most parsimonious model for explaining child wasting, balancing goodness of fit with model simplicity while adequately accounting for community-level variation.

Table 4 presents the results of a multilevel logistic regression analysis evaluating the associations between WASH and socio-demographic factors and child stunting. In the fully adjusted model (Model 3), several WASH factors emerged as significant predictors of nutritional status. Specifically, children living in households with improved latrine facilities had significantly lower odds of stunting compared to those using unimproved facilities (AOR = 0.693; 95% CI: 0.557–0.862; p = 0.001), highlighting a strong protective effect of sanitation. Access to water was also a significant factor, as children in households requiring less than 30 min to collect water had 1.39 times higher odds of stunting compared to those with water on the premises (AOR = 1.39; 95% CI: 1.11–1.75; p = 0.004). However, the type of drinking water source did not show a statistically significant association with stunting after full adjustment.

Regarding socio-demographic and geographic variables, the place of residence showed that children from nomadic backgrounds had significantly lower odds of stunting compared to urban residents (AOR = 0.716; 95% CI: 0.288–1.78; p = 0.001). Regional disparities were also prominent; children residing in the Sool (AOR = 1.69; 95% CI: 1.14–2.52; p = 0.007) and Sanaag (AOR = 1.60; 95% CI: 1.14–2.52; p = 0.013) regions experienced significantly higher odds of stunting compared to those in the Awdal region. Other factors, including child sex, age of the household head, household wealth quintile, and community-level poverty, were not found to be statistically significant predictors of stunting in the final model (p > 0.05). These findings suggest that improving sanitation infrastructure, reducing water-collection burdens, and targeting regional inequalities are essential strategies for reducing chronic undernutrition in children.

Table 5 presents a comparison of four multilevel logistic regression models for child stunting. Model 0 serves as a baseline, and Models 1–3 incorporate individual-, household-, and community-level predictors. Model fit improves across models, evidenced by decreasing AIC values from 4455.161 in Model 0 to 4432.278 in Model 3, indicating enhanced explanatory power. The log likelihood becomes less negative, reaching its peak in Model 3 (−2195.14). Although BIC values are lower for Model 0 due to penalization for added parameters, more complex models generally show higher BIC values. The intra-class correlation coefficient (ICC) remains around 20%, suggesting substantial variation in stunting between clusters. However, the ICC slightly decreases from 0.22 in Model 0 to 0.20 in Model 3, indicating that included factors explain some between-cluster variability. Cluster-level variance also decreases from 0.92 in Model 0 to 0.82 in Model 3, confirming that the predictors reduce unexplained community-level differences while still leaving some residual variation. In conclusion, Model 3 is the best-fitting model with the lowest AIC and highest log likelihood, validating the use of multilevel modeling for analyzing child stunting.

Based on the findings, the Somaliland Ministry of Health Development, in collaboration with UNICEF and other stakeholders, should prioritize interventions that address critical WASH deficiencies directly linked to child undernutrition. Given the strong association between longer water collection times and increased child wasting, immediate efforts should focus on improving access to safe drinking water sources within or very close to households, potentially through community water points, boreholes, or piped water systems. Additionally, the consistently high odds of wasting linked to unsafe child stool disposal demand targeted public health campaigns on proper child faeces management, including safe disposal practices and the importance of latrine use for young children. While the protective effect of open defecation against stunting found in this study is unexpected and warrants further qualitative and contextual investigation to understand underlying factors, the overall emphasis should remain on promoting improved sanitation facilities. This could involve community-led total sanitation (CLTS) approaches adapted to local cultural norms, alongside hygiene education programs emphasizing handwashing with soap at critical times, especially before food preparation and after defecation, to break the faecal-oral transmission routes that contribute to malnutrition. Finally, given the observed regional disparities and socioeconomic determinants, interventions should be geographically targeted to vulnerable regions like Sanaag and prioritize poorer households, integrating WASH improvements with broader maternal and child health and nutrition programs.