Although there have been a number of large-scale studies of public perceptions of drinking water safety in developed countries, this issue has been less widely studied in low and middle income countries, often through smaller scale studies . This study focused on a large-scale multi-year data set for South Africa, a middle income country but one with high levels of inequality. It confirms the finding from high income countries that consumers predominantly respond to organoleptic signs of contamination in their perception of drinking water safety, with socio-economic and demographic characteristics being comparatively less important. This is consistent with findings in other middle income countries, notably an urban Ukrainian population, which rated drinking water safety as fair to unsafe and also rated organoleptic properties as fair to poor .
Of the socio-economic and demographic characteristics considered, only ethnicity was included in the final model from the 2002 General Household Survey (Table 2). Despite the high levels of inequality, there was no evidence here that perception of drinking water safety differed by socio-economic status as measured through household expenditure. Of the three organoleptic properties, clarity and taste appeared more important than odour (Tables 2 and 3). However, despite this strong relationship, the link between organoleptic properties and perceived drinking water safety is likely to be complex. In a survey of tap water consumers in Cape Town, for example, nine out of 21 respondents complaining of poor taste cited too much chlorine as the cause . Similarly, five out of nine respondents (in formal residential areas) detecting an odour attributed this to chlorine.
There was some evidence that contextual clues – most notably the source type – were linked to perceived drinking water safety. There was also some evidence for a link between perceived control over water supplies and their perceived safety. In particular, in 2002, tapwater users relying on a public or neighbour’s tap were less likely to consider their water safe.
There has similarly been relatively little work exploring the change in consumer perceptions of drinking water safety over time. After controlling for the expansion in improved water supply access over the study period through the pooled survey data analysis, household perceptions of drinking water safety appear stable over time, although small changes in perceptions would be difficult to detect because of survey sampling errors. There was however an increase in the small number of consumers with water piped into the dwelling, who considered their water unsafe between 2003 and 2009 (Figure 1). A slight decline in such piped water that was free from organoleptic signs of contamination over this period is likely to account for the apparent trend in this group. This finding is perhaps unexpected, given the widespread cholera outbreak that affected South Africa in 2000–02 (Figure 5) and the link between prior history of water-borne disease and perceived drinking water safety . Elsewhere, a seasonal increase in household water treatment has been noted in Madagascar in response to a perceived increase in cholera risk , but there was no such discernible link here. This was despite a ministry of health campaign using radio, television and pamphlets warning of unsafe sources .
Other recent studies have used the GHS to examine other aspects of drinking and water resource management. The 2004 GHS has previously been used to explore attitudes towards environmental water pollution and uptake of home water treatment . Similarly, the 2002 GHS has been used to assess characteristics of households lacking access to safe water  as well as patterns of water supply cut-offs . Given that home water treatment involves households taking action to reduce consumption of unsafe water, this may provide additional insights into perceived drinking water safety.
In theory, it would be possible to compare the perceived safety of different source types with scientific measures of water quality from monitoring systems and other surveys of water supplies. Examples of such data include that published under the Blue Drop scheme  and earlier reports such as MacIntosh and Colvin  for groundwater or Ehlers et al.  for bottled water. However, such comparisons are likely to be problematic and were not attempted here because:
Many potential sources of water quality data are not restricted to the point-of-use, but also entail sampling of untreated source water or water within distribution systems.
Professional assessments of drinking water safety often rely on both testing of multiple water quality parameters and sanitary risk inspection. However, many data sources only include a subset of these parameters and may omit risk inspections altogether. Not only is there a problem of partial coverage of drinking water safety parameters in some data sets, but finding an appropriate way of integrating these measurements into a single composite metric of water safety (for example, by choosing the lowest water safety band across all measured parameters) is also potentially difficult.
Whilst there are monitoring systems and sample surveys available, few are designed to provide nationally or provincially representative estimates.
Since consumer perception of drinking water safety seems largely dependent on the organoleptic properties of water supplies, this implies that information about water quality disseminated through the media (such as through the South African Blue Drop reports) may not alter public opinion on safe water greatly. However, although the perceptions of the general public may be relatively intransigent, such campaigns may still be successful in mobilising specific community activists, consumer groups and in motivating water sector professionals to improve on service levels.
Because the study relied on a secondary data set, a number of potentially important influences on perceived drinking water safety were omitted from this analysis. Such influences include trust in water providers like municipalities; information exchange with neighbours and other contacts; previous experience of water borne diseases such as cholera; recent changes in the nature of drinking water provision; and household perceptions of risk . Unlike some other studies , we did not attempt to model the multiple inter-dependencies between potential influences on perceived drinking water safety, such as taste, clarity and source type. The potential for response bias  arising from the interaction between interviewer and interviewee in such government-sponsored surveys is also well documented, though the direction of any bias is unclear in the GHS. Although the surveys analysed covered many households, because of their inherent design, we were limited in our ability to identify specific locations where confidence in supplies was low, disaggregate results for specific source types, or detect slight changes in perceived water safety over time. More generally, the survey is useful for understanding general changes in public confidence in water supplies over time, but less useful in understanding the perceptions of specific sub-groups of the population who may be particularly important in holding service providers to account.
It is unclear how far South African media coverage of the water sector may have varied over the period of study or whether the relatively stable consumer perception of drinking water safety mirrors media coverage over the period. However, it seems likely that the cholera outbreak of 2000–02 in particular would have attracted substantial  media attention. However, media reports related to the water sector and outbreaks of diseases such as cholera require further investigation. There may also be scope in future research to examine specific sub-groups of the population such as those complaining directly to utilities  or awareness of specific campaigns to disseminate water quality information, such as the Blue Drop initiative.