Tap water quality: in the eye of the beholder

Tap water quality is important for health and well-being. Monitoring tap water quality is thus an essential public service. There is also an emphasis on the perception and level of satisfaction with tap water, which, for example, impacts the consumption of bottled water (Delpla et al. 2020). Several studies have investigated the relationship between perceptions and objective measures of tap water quality (e.g., Montenegro et al. 2009; Proulx et al. 2010), but the empirical evidence is patchy and thin.

This paper studies citizen satisfaction with tap water using a biannual nationwide citizen survey in Norway. By linking the individual responses to municipal-level test data supplied by Statistics Norway (SSB), we can investigate the link between objective water quality and user satisfaction.

The test data capture the share of inhabitants that are connected to water supply systems with acceptable levels of Escherichia coli, intestinal enterococci, pH, and color in a given year. Interestingly, only the test result for color is significantly associated with the citizens’ satisfaction with the tap water. Instead, a set of individual characteristics do a much better job in predicting how satisfied respondents are with their tap water.

Most important is a measure of the respondents’ ‘general satisfaction/mood’, a variable that captures the respondents’ average satisfaction with 10 other categories. This reflects that respondents tend to report high or low satisfaction consistently, across unrelated categories. We also consistently find that men report higher satisfaction with the tap water than women.

When looking at age-groups, we find a mixed bag of results. The oldest group (76 years and older) is significantly more satisfied with the tap water than any group of younger respondents. Satisfaction does not grow steadily with age, though. Rather, respondents in the youngest group (18–35 years) are equally satisfied as respondents in the second-oldest (56–65 years) group, while those in the groups between (36–45 and 46–55 years) are consistently less satisfied with the tap water.

Respondents with higher education are consistently less satisfied with tap water than those with lower or no education. Interestingly, we find an opposite effect for income. Respondents from households with a joint income of above 1 million NOK (about 100,000 USD), i.e., the 14% highest incomes in the sample, tend to be more satisfied with the tap water. Married (including those living with their partner) do not deviate systematically from un-married. Finally, we observe a weak tendency that respondents who voted for a left-leaning political party at the last national election before the survey are more satisfied with the tap water than others.

Several earlier studies have looked at factors explaining user perceptions of tap water quality, but unlike our study, these are based on modest datasets, mostly with little or no variation over time. Since we combine a national survey conducted every second year with annual testing data, we have data for well over 40,000 individuals in over 400 Norwegian municipalities spread over five surveys conducted over a period spanning a full decade.

The richness of our data does not only ensure better grounds for generalizability than earlier studies, but it also allows us to use refined empirical techniques. Since we have survey data for 5 years and multiple responders per municipality in each round of the survey, we can use fixed effects techniques to control for all aspects that are unique and time-invariant for a municipality (municipal fixed effects). Hence, our inference is based on different individuals drinking tap water within the same municipality. We can also control effectively for general trends using time fixed effects, securing that our results are not driven by, e.g., a general national improvement in water quality over time. In addition, we can combine these sets of fixed effects into municipality-specific time trends. When using this specification, the inference is based on individuals within the same municipality in the same year, i.e., all variables that vary over time and across local governments but are similar for all individuals within a local government in a particular year are controlled for. We do not include test results in this specification, as this is captured by the municipality-specific time trend, and focus only on the individual characteristics of the respondent.

Our results are consistent with findings in several other studies. Many studies find that flavor and other sensory qualities (odor, color) are important explanatory variables of tap water quality (Turgeon et al. 2004; Doria et al. 2009; Montenegro et al. 2009; Proulx et al. 2010; Piriou et al. 2015). These findings are consistent with our finding that color matters, as sensory qualities are directly observable to the consumers. Similar as us, the literature also frequently finds that many other, and less obvious, factors affect the assessment of water quality.

Socio-economic characteristics of the respondents are often important determinants. Indeed, Ochoo et al. (2017) found no correlation between actual water quality and citizen satisfaction in a study of 100 households in 45 communities in Newfoundland, Canada. They found that more highly educated and well-paid inhabitants approved of the water quality to a larger extent than less educated, low-income citizens. Delpla et al. (2020) studying perceptions of tap water among 1,014 citizens in Québec, Canada, found a weak link between quality and satisfaction with tap water. Water consumption, however, was strongly linked to sensory qualities, but also impacted by household water treatment (filtering and cooling), knowledge about water quality, and risk perceptions.

Also, earlier studies have investigated variables such as age, education, household income, and gender as predictors of water quality perception (Auslander & Langlois 1993; Turgeon et al. 2004; Dogaru et al. 2009; García-Rubio et al. 2016). All these studies are based on relatively small data sets. Romano & Masserini (2020) lists these and others, and the largest sample size in their list is around 1,000. Furthermore, some results are in conflict. For example, Dogaru et al. (2009) and García-Rubio et al. (2016) find no effect of gender, while Doria (2010) find that women express more concern and perceive risks with tap water as higher.

The importance of high-quality tap water is best illustrated by what happens when the quality is poor. Examples from Norway include an outbreak of waterborne giardiasis in Bergen in 2004 after contamination of a municipal water supply (Nygård et al. 2006), an outbreak of the gastrointestinal disease in Røros in 2007, sourced to groundwater waterworks (Jakopanec et al. 2008), and E. coli contamination of a water holding pool in Askøy in 2019 (Paruch et al. 2020). In the latter case, over 2,000 residents fell ill and two deaths were suspected to be associated with the contamination.

Our main source of data is the Norwegian citizen survey which is carried out every second year by the Norwegian Agency for Public and Financial Management. The survey is sent in fall in even-numbered years and runs until spring in odd-numbered years. Each survey is then thoroughly documented in reports published by the agency (Difi 2021). For ease of notation, we will refer to the year of the report when we date the surveys, i.e., we say the 2011 survey, rather than the 2010/11 survey and so on.

Among the questions are several that deal with how satisfied citizens are with a set of services, including tap water, and various questions that captures individual characteristics such as demographics and socio-economic background. Table 1 presents descriptive statistics for the citizen satisfaction with tap water in each year. In the survey, the scale runs from −3 to 3; we have transformed it to a 1–7 scale for practical convenience. Hence, 1–3 on our scale would indicate a negative assessment, 4 is neutral, whereas 5–7 is positive.

An important take-away is that citizens are mostly very satisfied with their tap water, and the responses are also quite stable over time. In the first (2011) and last (2019) surveys, 54% give tap water the best grade, while 58–59% do so in the surveys in the intervening years. Moreover, the share giving the second-best grade is consistently in the mid to high 20s, while the share giving the lowest grade is well below 1% in all surveys. Hence, we also see that mean is in the area 6.2–6.3 in all surveys. Despite the high satisfaction, there is still meaningful variation in the data that allows us to conduct empirical test procedures.

The owners of the water supply systems are required to report results to the Norwegian Food Safety Authority (Mattilsynet) each year in February. The data are then made available on SSB's webpages, where we collect them and then merge the data with the data from the citizen survey. Since the test data are reported in February each year, we use test data from the year the survey started, i.e., the even-numbered years. However, in order to keep years consistent across tables, we report the year the survey was completed (the odd-numbered years) consistently when we discuss the data. The variable SSB provides the share of the population that have access to water that satisfies a variety of test criteria, as listed in Table 2.

Table 3 summarizes the data for the test results. In the rightmost column, we see the averages for the dataset as a whole. These indicate that a large majority of Norwegian citizens have access to satisfactory tap water. To be clear, SSB defines water from a given water supply to be less than satisfying when more than 5% of tests from the supply breach the criteria in Table 2. For E. coli and intestinal enterococci, the average is well above 98%. For pH, the average is just over 95%, while for color it is close to 91%.

When looking at the year-by-year descriptive statistics, we see that the numbers are quite stable for all criteria except for color. While the averages for the other ones are consistently in the mid to high 90s, the average for color varies from a low 78.4% in 2011 to a high 99.5% in 2019. The trend is positive, the numbers grow for each observation.

Although the numbers are consistently high and quite stable over time for E. coli, intestinal enterococci, and pH, the standard deviations show us that there is a sizable variation in the data, especially for pH. Hence, it is possible to conduct meaningful empirical analyses with these variables. However, since there is much more variation in the test results for color than the other criteria, this is the variable that is most likely to produce significant results in the empirical analyses.

On the right-hand side, we first have the various test results, as discussed above, a variable that does not vary between citizens within the same municipality and thus do not have the i sub-script. We expect the coefficient to be positive, the share of the population with access to satisfactory drinking water should intuitively be positively correlated with user satisfaction.

The variable general mood is constructed as the average response to 10 other survey questions about how satisfied the respondent is with a variety of services and characteristics of the municipality¹. The variable captures that people vary systematically in how satisfied they are with society in general, and thus also the quality of the drinking water. Hence, we expect to come out with a positive sign.

The vector containing demographic variables include a dummy for whether the respondent is male, and a set of dummies splitting the respondents into age categories. Due to anonymity concerns, we do not have access to the respondents’ exact age. We do not have any strong ex-ante expectations of the sign of the coefficients for these variables.

The vector with socio-economic characteristics consists of several variables. First, we include a dummy equal to one if the respondent voted for a left-leaning party in the last national election before the survey took place. The Norwegian left is almost completely dominated by the social democratic Labor Party and a moderate democratic socialist party which frequently cooperates with labor. Second, we include a dummy equal to one if the respondent is either married or lives together with his or her partner. Third, we use a dummy equal to one if the respondent has a degree from a college or university. Fourth, we include a dummy equal to one if the respondent is part of a household with a joint income of 1 million NOK (about 100,000 USD) or more. As for the demographic variables, we have no particular expectations for the sign of the coefficients for these variables.

is a component that captures all variation in data that are unique to a municipality and constant over time. In the regression, we capture this using municipality dummies. This is called municipality fixed effects, and substantially decreases the potential for omitted variables as all time-invariant characteristics of a local government, such as geography or access and location of water sources, are effectively controlled for.

is a component that captures time trends that affect all municipalities and respondents in the same way. We capture these by including year dummies, i.e., time fixed effects. Since we have multiple respondents within municipalities in every survey, we can also combine the municipality and time fixed effects and construct municipality-specific time trends. When including these trends, the inference is based on individuals within the same municipality in the same year, i.e., we control for all variables that vary over time and across local governments but are similar for all individuals within a local government in a particular year. This control includes the test results for water quality, and tests are thus omitted from the regression in this specification. The specification is thus particularly well suited to investigate the effect of the variables describing the individual respondent. Note that we cannot use fixed effects on the individual respondent level, since we do not follow the same respondents over time.

Since our dependent variable is ordinal, we should also consider using an ordered probit model. As the number of categories in the dependent variable is rather large, the probit is not expected to make much of a difference. Hence, we favor the more simply interpreted OLS regression to the more technically refined ordered probit. Nonetheless, we will also report results from an ordered probit regression to show that the methods do not deviate substantially. Descriptive statistics for explanatory variables are provided in Table A1 in the Supplementary Material.

Our results are reported in Table 4. In columns (A)–(D), we introduce the different test criteria one by one. Column (E) contains the same variables as column (D), but is estimated using ordered probit. The coefficients in this column are thus not directly comparable to those in the other columns. The main take-away from this column is that our results do not hinge on the choice of estimation technique, since the results look very similar in terms of direction and statistical significance. In column (F), we introduce the municipality-specific time trends. Since these also capture test results, only the respondent-specific variables are included in this column. Since this is the specification that goes the furthest in controlling for other variables, this is likely the most reliable for studying how characteristics of the individual respondent affect the responses. The results are quite stable across specifications, though, so these additional controls do not seem to matter much for the results.

The first thing we note is the seemingly weak relationship between objective measures of water quality and citizen satisfaction. Of the four measures, three are far from reaching significance on any conventional level of significance, with coefficients that appear to be precisely estimated zeroes. Moreover, they also come out with an unexpected negative sign, although that is probably not worth putting much emphasis on.

Color does come out with the expected positive sign and the coefficient is also statistically different from zero. Despite the strong statistical significance, it is worth noting that the effect is of relatively modest size. An increase of one standard deviation in the share of the population with drinking water of a satisfactory color (27.36) is associated with an increase in satisfaction with the drinking water of about 0.05 or about 4.4% of the standard deviation for satisfaction (1.14).

It does make some sense from an intuitive perspective that color affects responses more than the other tests. While colored water is directly observable to consumers, marginal amounts of bacterium or deviations in pH are not. With a zero tolerance for both E. coli and intestinal enterococci, it is fully possible that deviations are reported, without the respondent ever being aware of it.

A second observation is that the general mood comes out with a strongly significant positive effect. An increase in the general mood by one standard deviation (0.88) is associated with an increase in satisfaction of about 0.38 or about 33% of the standard deviation for satisfaction. Hence, we see that the general mood explains a lot more of the variation in satisfaction with the drinking water than data for the objective quality.

When we turn to demographics, the first observation is that men are consistently more satisfied with the drinking water than women. A coefficient of about 0.05 translates into about 4.4% of the standard deviation for satisfaction. That is, the difference between men and women is the same as the effect of increasing the share of the population with water that satisfies the test criteria for color is very similar.

Second, we find some interesting age patterns. The oldest group (76 years and older) is significantly more satisfied with the drinking water than the younger groups. The difference is not only significant when comparing the group to the omitted control group (18–35 years), but also to any of the other age-groups. When compared to the control group, a coefficient of about 0.076 means the difference is in the area of 6.7% of the mean satisfaction. However, this does not mean that satisfaction improves steadily by age. Rather, both age-groups 36–45 and 46–55 years are significantly less satisfied with the drinking water than the youngest group. The difference for both groups compared to the youngest group is about −0.04, or approximately 3.5% of the standard deviation for satisfaction. The coefficient for the age-group 56–65 years is consistently negative, but smaller than for the two previous groups and mostly insignificant. For the age-group 66–75 years, the coefficient is positive, but consistently small and insignificant.

Hence, it seems the age effect takes the shape of a somewhat badly written U. The young are quite satisfied. The middle-aged are not so satisfied, the slightly old is in line with the young, and the seniors are even more satisfied. If we compare the groups that disagree most, age-groups 46–55, 76 years and older, the difference is around 0.12 or roughly 10% of the standard deviation for satisfaction.

The results for the socio-economic characteristics are a bit mixed. The coefficient for whether the respondent voted for a left-leaning party at the previous elections comes out as positive, but it is quite small and mostly insignificant. The richest model, which includes municipality-specific time trends, is borderline significant at the 10% significance level. The coefficient translates into about 1.7% of a standard deviation in satisfaction. Whether the respondent is married or not does not seem to affect the responses. The coefficient comes out as significant in all estimations but is never significant at any conventional level of significance.

Respondents with higher education are consistently less satisfied with the drinking water than those without a degree from a college or university. A coefficient of about −0.073 is equivalent to 6.4% of the standard deviation for satisfaction. For income, there is a trend that those in relatively high-income households are more satisfied with the drinking water than others. The coefficient is only borderline significant, though, and in some specification not even significant at the 10% level. The coefficient corresponds to about 2% of a standard deviation for satisfaction with the drinking water.

Inhabitants in Norway have access, almost always, to safe and high-quality tap water. It may have been somewhat colored earlier, but recently, color has also been very good. Satisfaction with tap water among Norwegian citizens, however, varies more, and somewhat unsettlingly, is not really related to objective measures of quality. Rather, satisfaction with tap water in Norway is mostly explained by the overall satisfaction with other things, such as schools, kindergartens, and museums. One may wonder if satisfaction with tap water impacts satisfaction with these factors. Age, education, income, political preference, and gender also factors into tap water satisfaction to various degrees. We can speculate on some mechanisms at play here. Some of those with higher education may be what we call connoisseurs or foodies and may be more sensitive to variations in quality. Also, household water treatment, for example with charcoal water filters, may be more common in high-income households.

One clear implication of our findings is that policy decisions related to tap water quality should not rely on how satisfied end users are with the tap water quality. Such decisions should rely on objective measures and other relevant aspects, such as socio-economic tradeoffs. Water supplies and water supply systems are costly to build and maintain. Despite the general affluence of the Norwegian society, parts of the public water supply systems are outdated. For example, subsequent to the outbreak of gastrointestinal disease in Røros, Norway, in 2007, several faults and problems with the distribution system were discovered, including 300 m of wooden pipes supplying the city center (Jakopanec et al. 2008). The wooden pipes were laid down in 1942.

All relevant data are available from an online repository or repositories. (https://dfo.no/sites/default/files/Fagomr%C3%A5der/Rapporter/2021/Innbyggerundersokelsen/Innbyggerundersokelsen-2021-Tabeller-med-alle-resultatene.xlsx and https://www.ssb.no/statbank/table/11786/).

The authors declare there is no conflict.