Self-reported anticipated harm from drinking water across 141 countries

Water crises are widespread and worsening, with impacts on agriculture, development, stability, and well-being. While service availability and access indicators are tracked globally (e.g., SDG 6.1), water quality data are incomplete and even safely managed sources can become contaminated between source and point-of-use. In the absence of trusted objective data, individuals’ perceptions and experiences often guide water-related decisions, yet perceptions are rarely measured systematically. The study aims to fill this gap by quantifying the prevalence of self-reported experienced and anticipated harm from drinking water across countries, identifying national-level predictors of perceived future harm, and determining individual-level characteristics associated with perceiving drinking water as harmful. The purpose is to highlight the role of user perceptions—shaped by trust, governance, organoleptic properties, knowledge, and values—in influencing behaviors that affect use, payment, and sustainability of water services, and to inform management and policy.

Prior work indicates that water users’ perceptions influence behaviors such as avoiding tap water, purchasing bottled water, and substituting sugar-sweetened beverages, with consequences for public health and system sustainability. Documented determinants of perceived water safety include taste and smell, institutional trust, knowledge of management and treatment, media access, risk awareness and tolerance, and personal values. Despite high levels of access to safely managed services in some high-income countries (e.g., U.S.), events like Flint’s lead contamination and frequent violations have eroded trust, increasing long-term bottled water use. Global indicators often miss point-of-use contamination and emerging contaminants (microplastics, PFAS). Perceptions and lived experiences have been shown to predict subsequent behaviors and expose inequalities by gender, urbanicity, and socioeconomic status. International bodies (WHO, UNICEF, World Bank) identify knowledge gaps in water quality as a barrier to global safety estimates, underscoring the need to measure perceptions alongside objective metrics.

Design and ethics: Secondary analysis of deidentified, nationally representative data from the Lloyd’s Register Foundation 2019 World Risk Poll (WRP), collected by Gallup with ethical approvals in each country. Verbal informed consent obtained; parental consent for minors where required.

Sample: Approximately 1,000 non-institutionalized individuals aged 15+ per country were surveyed in 2019 across 142 countries (total recruited n ≈ 154,195). Anticipated-harm data were missing for Kuwait; hence analyses of anticipated harm include 141 countries. Mode was telephone where coverage ≥80% or customary; otherwise face-to-face with multistage sampling. Post-stratification weights adjusted for nonresponse and demographics; projection weights used for pooled estimates.

Measures:

• Experienced harm: “Experienced or personally knew someone who experienced serious harm from drinking water in the past two years” (yes/no).
• Anticipated harm: Likelihood of experiencing “serious harm in the next two years” from drinking water with responses: not at all likely, somewhat likely, very likely. A binary indicator was created (somewhat/very likely vs not at all likely) to enhance cross-country comparability and address sparse extreme responses.
• Country-level covariates: Renewable freshwater resources per capita (AQUASTAT; log-transformed); percent of population with at least basic drinking water services (WHO/UNICEF JMP); percent wastewater treated (domestic/manufacturing; PANGAEA); percent of deaths attributable to unsafe water (GBD 2019; dichotomized <1% vs ≥1% based on lowess inflection); per capita GDP (USD, log-transformed) and World Bank income classification; Corruption Perceptions Index (CPI; Transparency International; 0–100 higher=less perceived corruption); national prevalence of self-reported experienced harm from drinking water (WRP).
• Individual-level covariates: Gender (man/woman; interviewer-assessed), self-reported difficulty getting by on present income (difficult vs getting by), urbanicity (large city/suburb vs rural/small town/village), education (≤8 years; 9–15 years; ≥4 years beyond high school).

Quality assurance: Cognitive testing, standardized translations (forward/back-translation), common risk definition read to all, interviewer training, audits of ≥30% face-to-face and 15% phone interviews.

Statistical analysis: National prevalence estimates computed with post-stratification weights; pooled/regional estimates used projection weights. Country-level analyses used weighted least squares (WLS) regressions with robust standard errors; observations weighted by inverse SEs of the dependent variable (national anticipated-harm estimates). Linear and quadratic terms examined; AIC guided functional form selection. A multivariable WLS model included all country-level covariates. Individual-level analyses used generalized linear models (binomial family, identity link) estimating prevalence differences (PDs), with country fixed effects to account for clustering. Effect modification by national income status tested via interactions and stratified models. Sensitivity analysis: multilevel mixed-effects logistic regression with country as random effect and individual/country covariates as fixed effects. Two-tailed tests, alpha=0.05; Stata 17.0 used.

Prevalence:

• Experienced harm: 14.3% (95% CI: 13.6%, 15.0%) reported personal or known serious harm from drinking water in prior two years (N=142 countries), ranging 0.9% (Singapore) to 54.3% (Zambia).
• Anticipated harm: 52.3% (95% CI: 51.2%, 53.4%) anticipated serious harm in next two years (N=141 countries), ranging 8.0% (Sweden) to 78.3% (Lebanon); data missing for Kuwait.

Country-level bivariate associations (WLS; Table 1; Fig. 2):

• Renewable freshwater resources (log): β=0.2 (95% CI: -1.5, 1.9); p=0.819; R²=0.001 (no association).
• Basic drinking water service coverage: nonlinear; linear β=3.3 (1.4, 5.2), quadratic β=-0.02 (-0.04, -0.01); p≤0.001; R²=0.234. Lower anticipated harm at the highest coverage, with heterogeneity (e.g., Greece 58.9% vs Finland 9.1% anticipated harm despite 100% coverage).
• Wastewater treated (%): nonlinear; linear β=0.02 (-0.2, 0.2), quadratic β=-0.004 (-0.006, -0.001); p=0.003; R²=0.366 (higher treatment generally associated with lower anticipated harm).
• E. coli contamination at point-of-use (23 countries): β=-0.1 (-0.3, 0.1); p=0.215 (no significant association; Fig. 3).
• Experienced harm (prior 2 years): strong curvilinear positive association; linear β=2.1 (1.6, 2.7), quadratic β=-0.03 (-0.04, -0.02); p<0.001; R²=0.529.
• Deaths attributable to unsafe water (≥1% vs <1%): β=12.6 (7.8, 17.3); p<0.001; R²=0.134 (higher anticipated harm where ≥1% of deaths attributable to unsafe water), though many low-mortality countries still had high anticipated harm (e.g., Lebanon ~0.1% deaths yet 78.3% anticipated harm).
• GDP per capita (log): nonlinear inverse; quadratic β=-2.7 (-3.8, -1.6); p<0.001; R²=0.473. High-income average anticipated harm 37.2% vs 54.8–57.1% in middle/low-income.
• Corruption Perceptions Index (0–100, higher=less corruption): nonlinear inverse; linear β=0.7 (0.1, 1.3), quadratic β=-0.01 (-0.02, -0.01); p≤0.02; R²=0.538 (largest explained variance). Example: Yemen (high perceived corruption) 52.3% vs Denmark (low corruption) 11.4%.

Country-level multivariable model (Table 2; N=130; R²=0.748):

• Experienced harm: linear β=1.5 (1.0, 2.1); p<0.001; quadratic β=-0.02 (-0.03, -0.01); p=0.001.
• Deaths attributable to unsafe water (≥1%): β=-5.9 (-11.5, -0.2); p=0.042 (direction reversed vs bivariate when adjusting for covariates).
• CPI: linear β=0.8 (0.3, 1.4); p=0.003; quadratic β=-0.01 (-0.02, -0.01); p<0.001. Predicted 16.6 percentage-point lower anticipated harm (95% CI: -27.5, -5.3; p=0.003) for CPI 80 vs 60, holding others constant.
• Other covariates (renewable water, basic service coverage, wastewater treatment, GDP) not significant after adjustment.

Individual-level predictors (Tables 3–4; Fig. 4):

• Pooled bivariate PDs: women +4.9 pp (95% CI: 3.4, 6.3); difficult to get by on income +4.1 pp (2.5, 5.6); urban +5.1 pp (3.1, 7.1); education 9–15 years +7.7 pp (6.1, 9.4); ≥4 years beyond high school +10.9 pp (8.4, 13.5); all p<0.001. Multivariable PDs similar in magnitude/direction.
• Effect modification by income level: no significant gender difference in low and lower-middle income settings; higher anticipated harm among women in upper-middle (+8.5 pp) and high-income (+10.5 pp) countries. Urbanicity associated with higher anticipated harm in lower-middle (+6.4 pp) and upper-middle (+6.5 pp) but not in low/high income. Higher education associated with greater anticipated harm across low, lower-middle, and upper-middle income strata; not significant for ≥4 years beyond high school in high-income countries.

Multilevel mixed-effects logistic regression (sensitivity): women had OR=1.25 (95% CI: 1.22, 1.28); individuals reporting experienced/known harm had OR=4.23 (4.07, 4.39) for anticipating harm; directions consistent with primary analyses.

The study shows that perceived risk of future harm from drinking water is high globally (52.3%), and varies in ways not fully explained by traditional supply-side indicators or objective quality measures. Associations with prior experiences of harm and with perceived public sector corruption suggest that trust and governance strongly shape risk perceptions. The lack of consistent alignment with renewable water availability and mixed relationships with infrastructure coverage and wastewater treatment highlight that service presence does not guarantee perceived safety, especially where contamination can occur post-collection or where governance is distrusted. Individual-level patterns—greater anticipated harm among women, urban residents, those struggling financially, and the more educated (context-dependent)—underscore social and contextual drivers of perceptions that can influence water use behaviors (e.g., avoidance of piped water, bottled water reliance, substitution with sugar-sweetened beverages). These findings address the research aims by quantifying the burden and determinants of perceived water harms, indicating that integrating user perceptions into monitoring and management is essential to improve uptake, payment, and sustainability of safe water services.

Anticipated harm from drinking water is widespread across diverse geographies and service levels, representing an underappreciated dimension of the global water crisis with implications for health and system sustainability. Incorporating user perspectives—particularly trust in safety and governance—into monitoring and policy can improve water security outcomes. Future research should: collect objective water quality and perception data concurrently; conduct longitudinal and sub-national analyses to identify causal pathways and context-specific drivers; examine the role of corruption in both public and private sectors; and evaluate interventions that build trust, enhance transparency, and address inequities (e.g., targeted communication, infrastructure improvements such as lead service line replacement, equitable access to safe alternatives).

Perceived anticipated harm does not necessarily equate to actual future harm; attributions of experienced/anticipated harm may reflect other environmental exposures (e.g., air or food pathways). Cross-sectional design provides a single time point, limiting causal inference and assessment of change; test–retest reliability was not evaluated. Potential self-selection and mode biases exist despite rigorous sampling and weighting; interview mode varied by income level and was not adjusted due to confounding. Some objective data were limited (e.g., E. coli contamination available in 23 countries), and residual confounding may persist. In country-level models, one specification (renewable freshwater) showed non-normal residuals. The Corruption Perceptions Index captures perceived public-sector corruption only, not private-sector governance relevant to water service provision. Collapsing Likert categories for anticipated harm, while improving cross-country comparability, may reduce granularity.