An anonymous postal questionnaire was administered to 7,500 rural households in six geographic regions of Saskatchewan in the fall of 2011. The questionnaire was distributed through Canada Post’s Unaddressed AdMail service, which provides delivery of bulk mail without specific addresses to houses and farms within a given postal code. Target postal codes were selected by using Canada Post Householder Counts in conjunction with postal code geography files [14]. A geographic information system (ArcMAP, ESRI, Redlands, CA, USA) was used to calculate the smallest radius around a central point selected for each region which would include the centroids of enough postal codes to encompass 1,250 eligible households. To ensure that the questionnaire would be distributed primarily to rural households, postal codes that did not include any farms were excluded, and where postal codes contained more than 200 houses, the survey was sent only to farms within that postal code.

Questionnaires were sent to 1,250 households from between nine and 12 postal codes (median = 10) in each of the six regions for a total of 60 postal codes. The resultant data included a multistage, hierarchical sample of respondents from households within postal codes selected from within each geographic region. As a result of this distribution process, the questionnaire was delivered to a sample of residents from 24% of the rural municipalities within Saskatchewan.

The four page survey consisted of questions about household water sources, perceptions of quality and health risks from drinking water, consumption of tap water and bottled water, home treatment of tap water, and demographics. The questionnaire was modified from one used in a pilot study in 2010. We requested that the questionnaire be filled out by one member of the household who was over the age of 18, and returned in a pre-addressed, postage-paid envelope. Distribution of reminders or follow up to households that did not respond was not practical given that the questionnaires were not addressed to specific households. Ethics approval was obtained from the University of Saskatchewan Behavioral Research Ethics Board.

The primary outcomes of interest were whether respondents reported primarily drinking bottled water and whether they consumed their household’s tap water daily. Of secondary interest was the choice to treat the household tap water in some way.

A causal diagram (Figure 1) was constructed to help guide the process of model development. Primarily choosing bottled water was recorded as a dichotomous variable based on a question about consumption of purchased bottled water in the home with three possible responses. If respondents chose “yes, it is the primary drinking water source” they were classified as primarily bottled water users, whereas those who chose “no” or “yes we drink it sometimes” they were classified as not using primarily bottled water. Daily consumption of tap water was also a dichotomous variable. Treating the tap water was evaluated as a dichotomous outcome based on the response to a question asking if the respondent had any equipment in their home to make the tap water better or safer to drink. Because the use of in-home treatment devices has been examined as a predictor of water consumption patterns in other studies [9,12,15], the use of water treatment in the home was also assessed as a risk factor in the models for bottled and tap water choices.

Risk factors examined included variables related to household tap water sources: use of a community managed water supply, use of a private water supply, and whether the water source is ground water or surface water. In rural areas, households sometimes use more than one water source; therefore, community and private supplies were not mutually exclusive, nor were ground and surface water sources, so each of these variables was analyzed separately.

Risk factors related to water quality and risk were also evaluated, including reporting any aesthetic complaint, the perception that tap water was not safe to drink, the fear that the water supply will become contaminated, and the perception that the tap water had made anyone ill. Reporting any aesthetic complaint was a dichotomous variable, recoded from a question for which respondents could select any number of choices from a list of complaints about their tap water. If any of odor, bad taste, discoloration or cloudiness were selected, the respondent was considered as having any aesthetic complaint about their tap water. The perception that the water was not safe to drink, fear that water would become contaminated, and the perception that the tap water had made someone ill were dichotomous variables and were based on questions for which yes or no responses could be given.

Whether or not the household had ever experienced a drinking water advisory was analyzed as a dichotomous risk factor. Respondents reported whether or not an advisory had ever been experienced, but not reasons for the advisories or the time frame within which past advisories were experienced. The number of years residing in the current community, age, gender, and whether there are children in the household were also analyzed as risk factors. Six age categories were recorded on the questionnaire (18–24, 25–34, 35–44, 45–54, 55–64 and ≥65 years); however, due to low numbers of responses in the three youngest age groups, these were collapsed into a single category, so that only four age categories were used in the analysis (i.e., 18–44, 45–54, 55–64 and ≥65 years). Four possible categories for the number of years residing in the community (0–5, 6–10, 11–20 and 21 or more years) were also collapsed into two categories (≤10 years, >10 years) for analysis.

Each of the outcomes was modeled using a generalized linear mixed model, specifying a binomial distribution and logit link function. Random intercepts were included in all models for both postal code and geographic region to account for any clustering arising from the hierarchical structure of the data.

Models were built for each outcome by first screening each risk factor individually where any risk factor with p < 0.2 was retained for consideration when building the final model. Manual backwards selection was used to build the final main-effects model, retaining only risk factors with p < 0.05. All risk factors dropped from the main effects model were then assessed for confounding based on whether its inclusion in the model led to a change greater than 10% in the regression coefficients for other risk factors. Biologically plausible two-way interactions between risk factors retained in the final model were assessed at a 0.05 level of significance; in the case of categorical variables, a type 3 likelihood ratio test was used to determine if the interaction was significant.

Models were built in Stata (StataCorp LP, College Station, TX, USA) with the xtmelogit command using a Laplacian approximation for efficiency. Using the risk factors identified in the model building process, the final model parameters were estimated with gllamm, using adaptive quadrature with 12 integration points. Estimates of the random effects and predicted probabilities for each model were produced using gllapred. Population averaged probabilities were estimated using the gllapred marginal function [16].

The proportion of the variance accounted for by postal code and region were examined for each of the outcomes, using an approximation of the variance partition coefficient for the binomial outcome based on the latent response variable model [17].

Values were missing for all outcomes and risk factors from at least one survey; any observations that were missing values for any of the risk factors or outcome for a given model were excluded from analyses including that variable. Therefore, the final number of observations used in each model varies and was reported for each model. Model assumptions were examined by evaluating the distribution of the residuals at the postal code and geographic region levels using Q-Q plots. The potential for outliers and influential data points was also investigated by plotting the standardized residuals at each level.

Of the 7,500 questionnaires sent out, 2,074 were returned. Seven were excluded because the postal code identifier had been removed by the respondent, one was excluded because it was blank, and one was excluded for being returned after the cut-off date for responses. As a result, 2,065 responses were used in the analyses, an effective response rate of 27.5%.

The median number of responses for each postal code was 44 (range 2–108) and the median number of responses per region was 353 (range 327–368). Frequencies were calculated for each outcome and risk factor (Table 1).

To assess the representativeness of our sample, the distribution of age categories and gender in our sample were compared to data from the Canada 2011 Census of Population [18] for the Census Subdivisions corresponding to the rural municipalities included in our survey regions (Table 2).

With respect to drinking water preferences, 30.8% of respondents reported primarily consuming bottled water (Table 1). Of the respondents with a private water supply, 30.7% (376/1,224) reported primarily choosing bottled water, while 28.1% (178/634) of those using a community water supply reported consuming primarily bottled water. Use of other types of water supply were less common; 32.5% (39/120) of those who used a public water station (i.e., a community-maintained, publically available fill station) and 39.5% (47/119) of those whose water was delivered by truck reported primarily using bottled water.

After accounting for other significant risk factors, reporting a water advisory increased the likelihood of choosing primarily bottled water (OR = 1.7, p < 0.001) compared to not reporting an advisory (Table 3, Figure 2). Of the respondents who reported ever having an advisory, 16.3% (79/485) of the respondents also reported having a current water advisory for their household. Of those that reported current advisories, 60.0% (42/70) reported drinking primarily bottled water and 34.8% (24/69) reported drinking their tap water daily.

Those who agreed that their tap water was not safe to drink were more likely to consume primarily bottled water than those who did not agree, but the effect of concern about unsafe tap water was greater for those who did not report an aesthetic complaints about the tap water (OR = 8.5, p < 0.001) than for those who did (OR = 2.3, p = 0.001) (Table 3, Figure 2). Compared to not reporting an aesthetic complaint, reporting any aesthetic complaint about tap water increased the likelihood of primarily consuming bottled water 6.1 times, but only for those who believed the tap water was safe (p < 0.001) (Table 3, Figure 2).

Those who lived in an area for ≤10 years were more likely to consume primarily bottled water than those who had not lived there as long (OR = 1.5, p = 0.01) (Table 3, Figure 3). Respondents who treated their tap water in their home and didn’t use a community water supply were more likely to primarily consume bottled water compared those who treated their tap water and used a community supply (OR = 2.5, p < 0.001) (Table 3, Figure 3). Compared to those who treated their tap water, respondents that did not treat their tap water were more likely to primarily consume bottled water, but the extent of the increase was greater for those who used a community water supply (OR = 4.6, p < 0.001) than for those who did not (OR = 2.5, p < 0.001) (Table 3, Figure 3).

The proportion of variance explained by postal code (4.3%) was greater than that explained by region (0.3%) in the final multivariable model for primarily choosing bottled water. This represents a 40% improvement in the variance explained by postal code compared to the random effects of the null model, in which postal code accounted for 7.1% of the variance and region accounted for 0.4%.

Most people (74.6%, (1,518/2,036)) reported consuming their tap water at least some of the time and 60.8% reported drinking tap water on a daily basis (Table 1). Of the respondents who used a private water supply, 63% (762/1216) reported consuming tap water daily. Daily tap water consumption was also reported by 61% (380/628) of respondents who used a community supply, 54% (63/117) of those who used truck-delivered water, and 55% (54/117) of those who used a public water station.

After accounting for other significant risk factors, reporting a water advisory decreased the likelihood of consuming tap water daily compared to not reporting an advisory (OR = 0.7, p = 0.004) (Table 4). Those who lived in an area >10 years were 1.6 times more likely to consume tap water daily (p < 0.001) than those who lived in an area for a shorter time (Table 4).

Those who did not think that their tap water was safe to drink were less likely to consume tap water daily than those who did, but the magnitude of the effect was slightly smaller for those who also reported aesthetic complaints (OR = 0.4, p < 0.001) than for those who did not (OR = 0.1, p < 0.001) (Table 4, Figure 4). When compared to those who believed their tap water was safe and had no aesthetic complaints, respondents who though their tap water was safe, but had at least one aesthetic complaint were 10 times less likely to consume their tap water daily (p < 0.001) (Table 4, Figure 4).

Reporting at least one aesthetic complaint also decreased the likelihood of consuming tap water daily for all respondents, with the effect of an aesthetic effect being greater for those who did not treat their tap water (OR = 0.1, p < 0.001) than for those who did (OR = 0.2, p < 0.001) (Table 4, Figure 4). Those who treated their tap water in some way were more likely to consume tap water daily compared to those who did not, but the importance of treatment was greater for those who also reported an aesthetic complaint (OR = 3.7, p < 0.001) than for those who had no complaints (OR = 1.8, p < 0.001) (Table 4, Figure 4).

The proportion of the variance in the final multivariable model for consuming tap water daily explained by postal code was 3.1%, while a negligible proportion was explained by region of the province (<0.001%). This represents a slight improvement compared to the random effects in the null model, in which postal code accounted for 4.9% of the variance and region accounted for 0.4%. The main effects in this model explained nearly 100% of the already small variation between regions of the province, and 38% of the variation between postal codes.

Of the respondents, 47.6% reported that they treat their household tap water in some way (Table 1). Of the respondents using a private water supply, 52.4% (637/1,215) reported treating their tap water, while 42.5% (264/621) of those who used a community water source reported treating their tap water in the home. Of those who used a surface water source, 41.6% (248/596) treated their water, while 50.7% (666/1,314) of those who used a ground water source reported treating their tap water. Of the respondents who indicated which type of treatment they used, 58.7% (501/853) used a water softener, including 68.2% (416/610) of those using a ground water source and 34.1 (76/233) of those using a surface water source. Of all respondents using water softeners, 72.1% (361/501) also indicated that some other form of water treatment (e.g., reverse osmosis, jug filter, ultraviolet, and distillation) was used.

After accounting for other risk factors, those who used a private supply were more likely (OR = 2.1, p < 0.001) to treat their household water compared to those who did not use a private supply (Table 5, Figure 5). Those who reported their tap water was not safe were half as likely to treat their tap water (p < 0.001) (Table 5, Figure 5) than those who did not. Having children under 18 residing in the home increased the likelihood of treating the tap water (OR = 1.6, p < 0.001) compared to not having children in the home (Table 5, Figure 5). Estimates were adjusted to minimize potential confounding by whether the home was in a town, and whether or not a community water supply was used.

The random effects explained only a small proportion in the variation between postal code (2.0%) and region (1.1%) in the final multivariable model with little change from the proportions in the null model (2% and 1.5% respectively). This suggested there were few differences among postal codes and regions regarding the decision to treat water.

The risk factors identified for consuming primarily bottled water, consuming tap water daily and choosing to treat tap water were summarized in Table 6.

Just over 30% of our respondents reported using primarily bottled water, with little difference between those who used a community water supply and those who used a private water supply. A recent national Canadian survey [1] reported that 20% of all Saskatchewan respondents reported consuming primarily bottled water, and that 19% of respondents using a municipal water supply and 39% of respondents using non-municipal supplies reported using primarily bottled water. However, 93% of the Saskatchewan respondents to the Statistics Canada survey [1] reported using a municipal water supply, compared to just 31% of respondents in our rural study population.

We did not attempt to quantify the proportion of bottled water consumed that would define a respondent as choosing primarily bottled water; whereas, some water consumption studies have set a threshold of greater than 75% bottled water use (e.g., [8,9]). Other Canadian studies had rates of primary bottled water use ranging from 22% [1] to 35% [19]. A recent national US study reported that just 13% of respondents reported using bottled water [4].

Among rural Saskatchewan residents, the choice to consume primarily bottled water at home appears to be mediated by a number of related factors. An interaction was identified between the belief that the tap water is not safe to drink and whether the respondent had any aesthetic complaints about the tap water. Having any aesthetic complaint increased the likelihood of choosing primarily bottled water, but only for respondents who felt their tap water was safe. Those who felt their tap water was not safe to drink were consistently more likely to choose bottled water than those who felt it was safe regardless of aesthetic concerns.

Although this interaction between these risk factors is unique to the present study, our findings build on previous studies which reported that bottled water use was related to aesthetic complaints [6,10,11,19] and perception of health risks from tap water [4,5,6,7,10,19,20].

Another interaction was identified between use of a community water source and whether respondents use some sort of in home treatment for their tap water. People who used a community water supply compared to some other supply were less likely to choose bottled water, but this was only true for those who treated their tap water. Respondents that treated their tap water were also less likely to consume primarily bottled water than those who did not, but the difference was greater for those that used a community water supply. Home treatment was identified as a factor that reduces the likelihood of consuming primarily bottled water in previous studies [8,9]. Its interaction with the use of a community water supply might reflect the use of home treatment devices to remove chlorine taste from tap water [9].

Respondents who reported having a water advisory were more likely to primarily consume bottled water. To our knowledge this has not been previously investigated as a risk factor. Drinking water advisories in Saskatchewan are issued for a variety of reasons. For larger distribution systems (flow rate > 18,000 L/day flow) precautionary advisories were most commonly issued in the year prior to our survey for operational reasons such as depressurization of the system, line breaks, planned maintenance or high turbidity levels which could compromise treatment, while in smaller systems precautionary advisories more often resulted from positive bacteriological testing. In both types of systems, emergency boil water orders were most often due to finding coliform bacteria in the water [21]. It is possible that having experienced an advisory could introduce doubts about the safety of the household water. Because those respondents reporting an advisory included some respondents that were currently under an advisory, the possibility that being under a current advisory was driving the consumption of bottled water was considered. However, relatively few respondents (16%) who reported ever having a water advisory also reported a current water advisory for their household. Furthermore, of the respondents with a current advisory, over a third reported drinking their tap water daily. Although this suggests that respondents might drink their tap water despite being under an advisory, it is also possible that respondents reported their typical daily beverage choices as opposed to the choices made specifically during an advisory situation. Given the importance of experiences with water advisories in choices made around drinking water the impact of water advisories on drinking water habits should be studied in more depth.

Respondents that had lived in the area for longer than 10 years were less likely to choose bottled water. Familiarity with the tap water has been identified as an important factor in perception of water quality [22,23], and this result suggests that familiarity also reduces the likelihood that alternatives to tap water will be sought.

The role of demographic variables such as age and gender has been inconsistent in previous studies [3]. Although age was unconditionally associated with primarily choosing bottled water in the present study, age was not included in our final model for primarily consuming bottled water. Some studies have reported that age as a significant predictor of bottled water usage [4,8,9] with the consumption of bottled water declining above age 30 in at least two studies. The population of respondents to our questionnaire was skewed toward older age groups which could have reduced the power of our study to detect differences between older and younger age groups. However, it is also possible that the effects of age are mediated through other risk factors included in our model.

The role of gender is less clear. In some previous analyses that examined the effects of gender in isolation, being female was associated with increased consumption of bottled water [6,10]. However, using multivariable analysis, Dupont [10] reported that males with children were more likely to be bottled water users than males without children, while Hu [4] reported that females had increased odds of drinking bottled water. However, gender was not a significant risk factor in another study [5] or in the present study. It is possible the effect of gender depends on other factors in the population under study or its effects are partly mediated through other risk factors such as perceptions of quality and risk.

More than 60% of our respondents reported drinking their tap water on a daily basis. We did not clarify if these respondents were primarily consuming tap water, but just 3% (40/1,213) of the respondents in this group also indicated that they primarily consume bottled water. Consequently, daily consumption of tap water was the only measure we had available to classify respondents as regular users of tap water. Considering that some respondents may not typically consume water on a daily basis, we could have underestimated the number of respondents who primarily consume tap water. In a national Canadian survey [1], 76% of Saskatchewan residents reported consuming primarily tap water. In the same survey, 78% of residents on a municipal supply reported drinking tap water, while only 49% of those with private water supplies chose primarily tap water. Among our respondents the proportion of respondents that reported drinking tap water daily was similar among users of private and community supplies.

The groups of risk factors included in the final model for choosing to drink tap water daily were similar, though not identical, to those for choosing primarily bottled water, but with opposite effects. Survey participants that reported water advisories were less likely to report daily tap water consumption, suggesting that this experience might reduce their confidence in the safety of tap water. Living in an area for longer than 10 years was perhaps a measure of familiarity with the water, and increased the likelihood of consuming tap water daily.

Reporting any aesthetic complaint decreased the likelihood of daily tap water consumption, but its effect was modified by both whether residents believed the tap water was safe, and whether residents treated their tap water; these interactions appeared to be independent of each other.

Reporting an aesthetic complaint made it less likely that tap water would be consumed regularly, but only when the tap water was considered safe. The belief that the tap water was not safe to drink made it less likely that tap water would be chosen regardless of the presence of any aesthetic complaints. This interaction was the inverse of a similar interaction found in the model for choosing to consume primarily bottled water.

The effect of reporting an aesthetic complaint on the likelihood of consuming the tap water was somewhat mitigated by treating the tap water, and treating the water had a greater effect on the likelihood of consuming tap water when an aesthetic complaint was reported. This interaction was important to the likelihood of choosing to consume tap water daily; whereas, having an aesthetic complaint was not an important risk factor for the decision to treat the water on its own. It appears that aesthetic qualities are important to the decision to treat only in to the context of whether the tap water is consumed regularly.

The choice to treat tap water was evaluated as a risk factor for the consumption of tap water even though the direction of the causal relationship between treating tap water and drinking tap water is not clear. For instance, the tap water might be chosen because of the perception that treatment has made it more safe or palatable, or the decision to treat might be made if tap water is the only viable option for drinking water and it is perceived to not be safe or palatable unless treated.

Few previous studies have examined the risk factors associated with primarily choosing to drink tap water. Dupont et al. [10] used analysis of variance to examine factors associated with the proportion of tap water consumed relative to filtered and bottled water in Canada, and found that the degree of concern about health risks from tap water was inversely related to the proportion of tap water consumed, as was the presence of various aesthetic concerns. This was similar to our results and underscores the importance of perception of quality and risk in making choices about drinking water.

Given the similarity between the models for primarily consuming bottled water and regularly consuming tap water, it might be reasonable to assume that similar factors, acting in opposite directions, play a role in each choice. However, there were some differences in the risk factors for each choice, and it has been hypothesized that choosing bottled water is not necessarily an alternative to choosing tap water, but may instead be considered an alternative to other pre-packaged beverages such as soda and juice [3,24]. The relative importance of bottled water as an alternative to tap water compared to other beverages requires further investigation.

We also investigated the risk factors associated with the decision to treat the household tap water. Use of in-home water treatment devices has become common [25]. Statistics Canada [1] reported that 50% of Saskatchewan residents indicated that they treat their tap water with a purifier, filter, or by boiling prior to consumption. More than 47% of our respondents indicated that they used any type of equipment in the home intended to make the tap water “better or safer to drink.” This number includes respondents that used water softeners, which are not recommended for the treatment of drinking water [25]. In other Canadian studies, water softeners were included as treatment devices, and rates of water treatment were similar [8,9]. In the present study, of those who indicated the type of treatment device used, 72% of respondents who used a water softener also used another device intended to treat drinking water.

Our results indicated that believing the tap water was not safe reduced the likelihood of treating the tap water. This is contradictory to another cross-Canada study that found health concerns increased the likelihood of consuming filtered tap water [10]. However, there could be a substantial difference between having general health concerns about water and the belief that the water is not safe. It is possible that if respondents felt their tap water was unsafe, they had no intention of consuming the water so did not treat it, or did not trust that home treatment devices would make their water safe.

Use of a private water supply increased the likelihood that water would be treated, a finding opposite to a study in British Columbia that reported fewer private source users than expected treated their tap water in an unconditional analysis [9]. This discrepancy may be related to the high rate of use of water softeners among our private water supply users. A study in Nova Scotia also reported that respondents with private water supplies were less likely to treat their water than those connected to a municipal supply [26]. However, in that study the use of bottled water was considered a type of water treatment making it difficult to directly compare their results with the present study, where bottled water consumption was considered separate from treatment.

We did not find an association between perception of poor aesthetic quality and the decision to treat the tap water among our respondents. This finding contradicted the study performed in Nova Scotia which found that treating household water was associated with the perception of lower water quality [26]. However, as previously mentioned, the risk factors identified in the Nova Scotia study may differ from ours because we evaluated bottle water consumption separately from other types of treatment.

The presence of children in the home increased the likelihood that the tap water would be treated. Dupont et al. [10] found a similar relationship but only for males, whereas gender was not included in our final model. Our model did include confounding variables, suggesting that the factors leading to water treatment are complex and deserving of further study, especially with respect to clarifying the factors related to treatment intended to make drinking water safer or more palatable compared to addressing the mineral content of the water.

As previously discussed, our models for tap water and treatment were limited by self-reported measures of relative tap water consumption and the goals of treatment. It would have been ideal to be able to model the risk factors for choosing primarily tap water for comparison to choosing primarily bottled water, rather than comparison to drinking their tap water on a daily basis. Previous studies have suggested that water consumption decreases with age [8,9,15], which could make daily tap water consumption an especially poor proxy for choosing primarily tap water in older age groups. Overall, the purposive nature of our regional sampling and a relatively low response rate, especially among younger age groups (Table 2), might limit the generalizability of our findings.