Rural Ghanaians Judge Water Risk Well in Broad Strokes, but Household-Level Calibration Falters

When a family in rural Ghana fills a bucket from the local borehole, the decision about whether that water is safe to drink rests largely on judgment rather than on laboratory analysis. A new study from the Nadowli-Kaleo District in Ghana’s Upper West Region suggests that this judgment is more sophisticated than critics of perception-based assessments might expect, yet still too imprecise to be trusted on its own. The research, published in PLOS Water, examined how closely what households believe about their drinking water matches what trained observers can detect about contamination risk, and the answer reveals a striking split between collective accuracy and individual error.

The study surveyed 425 households in a cross-sectional design, pairing each household’s subjective rating of its drinking water quality with an objective field assessment. Perceived quality was captured on a standard five-point Likert scale, ranging from very poor to very good. The objective side of the comparison came from a structured observation checklist that recorded visible and situational indicators of sanitary risk around water sources and storage, which the researchers aggregated into an ordinal risk index running from zero to four. This index serves as a proxy for contamination rather than a direct microbiological measurement, a distinction the authors are careful to emphasize, but it provides a consistent, field-ready gauge of the conditions that typically produce unsafe water.

At the population level, the correspondence between perception and observed risk was remarkably strong. The researchers computed Spearman’s rank correlation between the two measures and found a coefficient of negative 0.780, statistically significant at well below the 0.001 level. Because perceived quality was scored in the opposite direction to contamination risk, the negative sign is exactly what good calibration would predict: households that rated their water highly tended to face lower observed sanitary risk, and households that expressed doubt tended to face higher risk. In other words, the community as a whole had developed a shared sense of which water sources and practices were dangerous, and that sense tracked the observable risk gradient with impressive fidelity.

That population-level success, however, masks considerable noise at the level where decisions are actually made: the individual household. The researchers defined a household as calibrated when the absolute gap between its perceived quality score and its reverse-scaled observed risk index was 0.5 or smaller. By this criterion, only 49.4 percent of households qualified, and the mean gap across the sample was negative 0.86, indicating a systematic tendency toward misalignment rather than random scatter. The choice of threshold mattered substantially: tightening it to plus or minus 0.3 dropped the calibrated share to 26.1 percent, while loosening it to plus or minus 1.0 raised the figure to 63.1 percent. The authors describe this pattern as an imperfect calibration gradient, a phrase that captures both the genuine signal in perceptions and the substantial residual uncertainty that surrounds it.

What explains why some households read their water risk accurately while others do not? The most intuitive candidates are the standard socioeconomic variables that public health research often reaches for first. Education, in particular, has long been assumed to sharpen risk appraisal, on the theory that schooling equips people with the scientific literacy to recognize contamination hazards. The data told a different story. Among households with no formal education, 55.5 percent were calibrated; among those with secondary education, 44.1 percent; and among the small postgraduate group, only 12.5 percent, though that last figure rests on just eight households. A Pearson chi-square test across education categories yielded a statistic of 7.24 with four degrees of freedom and a p-value of 0.124, which fails conventional significance thresholds. The apparent downward trend with education, in short, cannot be reliably distinguished from sampling noise.

The researchers went further, testing whether the relationship between education and calibration might be non-linear, following an inverted U-shape in which moderate education helps but very high education somehow distorts judgment. A quadratic logistic regression produced an odds ratio of 0.94 for the quadratic term with a p-value of 0.462, offering no support for such a curve. Income and duration of residence in the community were likewise not significant predictors of calibration accuracy. This null result carries practical weight: it means that programs cannot simply target households by wealth, schooling, or tenure and assume they have identified the people whose water-safety perceptions are trustworthy or in need of correction.

The findings arrive at a moment when global water policy is leaning heavily on household-reported data. Under Sustainable Development Goal 6, which commits countries to safely managed drinking water for all, progress is monitored substantially through surveys that ask households whether their water is accessible, available when needed, and free from contamination. If perceptions were well calibrated at the individual level, such self-reports would be an efficient and inexpensive surveillance tool. The Ghana study suggests a more cautious interpretation. Because perceptions track broad risk gradients well, perception data may be useful for mapping relative risk across communities and identifying which water sources are widely regarded as problematic. But because nearly half of households fall outside even a generous calibration band, perception-based metrics alone appear insufficient for individual-level risk assessment, and treating a household’s own report as a reliable verdict on its water safety could leave genuinely contaminated supplies unexamined.

The methodological design deserves attention as well. The observed risk index was built from a structured checklist of sanitary conditions rather than from laboratory testing for fecal indicator bacteria, so it measures the conditions that epidemiological experience associates with contamination, not contamination itself. This choice reflects the realities of fieldwork in low-resource settings, where microbiological analysis of hundreds of household samples is often logistically and financially prohibitive. It also means the study’s conclusions concern the alignment between perception and observable sanitary risk, a narrower but still policy-relevant construct. Future work combining perception surveys with direct water-quality testing would help establish whether the calibration gradient documented here extends to microbiological outcomes, and whether the systematic mean gap of negative 0.86 reflects optimism, pessimism, or a mismatch in how the two scales capture the same underlying reality.

The study’s implications for risk communication are perhaps its most actionable contribution. If education, income, and residence duration do not predict who misjudges their water risk, then interventions cannot be efficiently targeted by demographic profile. Instead, the authors argue for objective monitoring paired with tailored risk communication, an approach in which water-quality testing identifies genuinely hazardous supplies and messaging is designed to correct specific misperceptions rather than broadcast generic warnings. Such tailored strategies acknowledge that perception and reality are correlated but imperfectly aligned, and that closing the gap for the roughly half of households outside the calibration band requires information they do not currently possess. In a district where most families rely on sources they assess by eye, by taste, and by reputation, the study offers a sobering quantification of how far reputation can drift from reality, even in a community whose collective instincts are, on average, remarkably sound.

For the broader field of water and sanitation research, the study adds a methodological caution and a conceptual framework. The sensitivity of the calibrated-household proportion to the threshold chosen, from 26.1 percent to 63.1 percent across a range of gap definitions, illustrates how sensitive calibration statistics are to arbitrary analytical choices, and the authors’ transparency about this range is a model for future studies of perception accuracy. The imperfect calibration gradient itself, strong in the aggregate yet noisy at the individual level, may prove to be a general feature of environmental risk perception in settings where formal monitoring is scarce: communities learn the statistical shape of their hazards through accumulated experience, but individual households lack the feedback needed to locate themselves precisely within that shape. Bridging that last mile, between a community’s collective wisdom and a family’s specific decision about the water in its bucket, is likely to require the objective measurement infrastructure that Sustainable Development Goal 6 envisions, delivered in forms that local institutions can sustain.

Subject of Research: Concordance between perceived drinking-water safety and observed sanitary contamination risk in rural Ghana

Article Title: Perceived drinking-water safety and observed contamination risk: Evidence of an imperfect calibration gradient in rural Ghana

Article References: Alhassan, A., Boyeh, Z., & Yahaya, A.-K. (2026). Perceived drinking-water safety and observed contamination risk: Evidence of an imperfect calibration gradient in rural Ghana. PLOS Water, 5(9), e0000590. https://doi.org/10.1371/journal.pwat.0000590

Image Credits: AI Generated

DOI: 10.1371/journal.pwat.0000590

Keywords: drinking water safety, water quality perception, rural Ghana, sanitary risk, risk calibration, SDG 6, household survey, water monitoring, risk communication, PLOS Water, Upper West Region, public health