Despite a high drinkable quality, many people avoid tap water because of vague anxiety about its safety. Conjoint analysis (CA) was conducted to determine what factors are considered important for consumers’ selection of drinking water. The information provision effect was also investigated inside CA profiles using different model equations. Results indicate that the perception of the safety of tap water was much lower than that of other waters. Higher levels of water hardness and cancer risk negatively influenced selection of drinking water, while third-party certifications about taste and safety positively impacted it. When cancer risk was shown in a CA profile, the weight given to other attributes decreased. Among different socio-demographic groups, gender was important in establishing drinking water preferences with men paying less attention to the benefits of water-dispensers and certifications from third parties. Besides, age also has some influence on drinking water selection. People's consciousness of taste, safety, cost, and handling for drinking water were assessed using an analytic hierarchy process and the scores were incorporated in a CA equation. The results suggest that improving people's perceptions of the taste and safety of tap water can promote consumers’ selection of tap water as drinking water.

INTRODUCTION

People's selection of drinking water is highly dependent on the hygienic condition of their living area. In some areas, especially in developing countries, bottled water is the most selected preference because of the poor tap water quality. However, even in areas where the drinkable quality of tap water is guaranteed, some people prefer bottled water, not only because of the taste, but also because of a vague feeling of anxiety surrounding tap water safety.

The Tokyo Metropolitan Government Bureau of Waterworks (TBW) has promoted a campaign for advertising tap water called the ‘Safe and tasty water project’ since 2004. Advanced water treatment techniques such as ozonation have been widely introduced, and tap water quality has attained the highest level in the world. Nevertheless, a consumer report revealed that the proportion of people who usually drink tap water is only at 49% in Tokyo so far (TBW 2013). In addition to the water's taste, feelings of anxiety about its safety are one of the reasons why people do not drink the tap water. Several other consumer reports in Japan (Suntory Co. 2013; Mitsukan Co. 2013) also provide similar reasons why people do not drink tap water. In the TBW (2013) report, more than 40% of people answered ‘no’ in response to the question, ‘Do you think the tap water is safe?’ This gap between the high quality of tap water and consumers’ perception of it indicates poor communication between water suppliers and consumers.

In Tokyo, where only 43.1% of people know that tap water is strictly regulated and must pass a higher number of tests than bottled water (TBW 2013), information provision is one of the possible measures to enhance people's perception of tap water. There are many selection criteria for choosing drinking water, including risk/safety features such as cancer risk, taste, and price. Therefore, it is important to know which features in addition to risk features have a significant effect on people's selection of drinking water. Besides, the expression of risk, whether by comparison with standard risk values or expressed as absolute values, can also have an impact on people's perceptions.

The selection of drinking water is based on ‘total utility’, which represents a person's total satisfaction with the target water. The total utility consists of ‘partial utilities’, which represent the person's satisfaction with each component of water, such as taste, safety, and price. Conjoint analysis (CA) is the analytical tool employed to evaluate partial utility independently. Alriksson & Öberg (2008) summarize the CA studies conducted for environmental evaluation, where only one of the 84 environmental studies they collected was about risk. Outside of the environmental field, there have been some trials of risk studies using CA. Baker & Burnham (2001) analyze people's preferences regarding genetically modified organisms (GMOs) by using corn flakes as an example. Their work revealed that people who have lower risk perceptions of GMOs and higher levels of knowledge about biotechnology are more likely to accept GMO foods. Although some studies have used the risk aspect as one of the attributes of the CA, information provision and risk expression effects inside the CA profiles have not been well investigated.

This study aims to reveal which features of and information about drinking water have more impact on its selection in the CA framework, and focuses on risk information provision and risk expression. The outcome can help water suppliers know what kind of information and risk expression should be provided for better communication with consumers. Furthermore, this study will provide a new approach to the use of the CA framework for analysis of the effect of information provision. Usually information is provided before the questions for CA; however, in this study, we provided information inside the CA questions and analyzed the effects on people's drinking water selection by using CA.

MATERIALS AND METHODS

Research framework

We focused on four types of drinking water: tap water, filtered water, bottled water, and water provided by a water-dispenser, and it is assumed that the total utility of drinking water consists of the following partial utilities: ideas about the water type, taste, risk, and certification.

It can be assumed that people have preconceived ideas about water based solely on its type, whether it is tap water or not. As shown in Jacoby et al. (1971), brand image can be one of the influential factors that determine the evaluation of a product's quality. Grubb & Hupp (1968) also report that people's beliefs about a product can significantly influence their product selection. For example, when water is shown as ‘bottled water’, a person has some good or bad image of the water from the name and it can influence his/her selection of the water. Therefore, people's ideas about the target water type can be one of the influential variables of water selection and a partial utility of drinking water.

Taste and risk evaluations influence people's drinking water selection. To determine how people evaluate water taste, the categories of ‘hardness’ and ‘remaining chlorine’ are important (de França Doria et al. 2009). Remaining chlorine in the tap water often leads to a poor evaluation of the taste; however, it is assumed to be already involved in the brand image of tap water. The risk in drinking water, especially in developed countries, mainly comes from the cancer risk caused by disinfection byproducts such as trihalomethane and haloacetic acids, while the main risk in developing countries comes from pathogens. In this study, we focus on the citizens in Tokyo; therefore, cancer risk is involved as one of the partial utilities of drinking water.

In addition, it is assumed that certification about the taste or safety of water by third parties can increase the total utility of the drinking water. Laric & Sarel (2008) examined the certification mark effects on consumers’ perceptions and concluded that most respondents believed that a product with a certification mark was better than products without it. Dimara & Skuras (2003) also showed that people were more likely to select a wine with certification. On the other hand, Lee & Turban (2001) reported that there was no significant influence of third-party certification on internet shopping. As shown here, the effect of the third-party certification on product selection would depend on the product types. There is no certification program for drinking water; however, it seems that some people would make their decision based on the mere presence of a certification mark. Therefore, we decided to evaluate whether certification information has an impact on drinking water selection or not.

In most cases of CA, the cost aspect is usually involved as one of the partial utilities; however, we assume that the cost aspect is to some extent already involved in the beliefs held or the image of each water type. Therefore, it is excluded from the utility function itself, and the effect of cost information is evaluated separately from CA profiles.

The basic concept of CA is shown in Equation (1) below, where the total utility of option j in person i (Uij) consists of observable total utility (Vij) and unobserved error term (ɛij). 
formula
1
Based on the above concept for CA, the basic function of Vij in this study is described as Equation (2). This is the base model as we will discuss later. 
formula
2
The example used in this study is shown in Figure 1. For the person i, if the person's total satisfaction (U) for the left option is larger than the right option, he or she will select the left option. V is a summation of individual satisfaction (partial utility) for each aspect shown in the figure: ‘Type of water’, ‘Hardness’, ‘Cancer Risk’, ‘Certification of Taste’, and ‘Certification of Safety’. The satisfaction for each aspect consists of the person's preference for the aspect (β) and the value of the aspect (x). For example, the person's satisfaction for ‘Cancer risk’ of the left option is determined by his/her preference on ‘Cancer risk’ and the value shown for the ‘Cancer risk’ in the left option (1/100,000 in this case).
Figure 1

Example of pair-wise choice set for CA questions. O: there is certification, -: no certification.

Figure 1

Example of pair-wise choice set for CA questions. O: there is certification, -: no certification.

As explained above, in Equation (2), k is each aspect (‘attribute’) of drinking water. The parenthetic part (xjk) represents the value of k in option j. βk represents the preference of person i on the attribute of k. Then, βikxjk is the partial utility of k for person i. The first three terms in Equation (2) represent the utilities derived from the consumers’ beliefs about the water types. If the water type in option j is tap water/bottled water/water-dispenser water, [TAP/BT/WD] becomes 1. To avoid multicollinearity, one of the water types is excluded from the equation. HDN is hardness and CNR is cancer risk, which expresses the probability of an increased number of people developing cancer as a result of drinking this water during their lifetime. CT and CS, respectively, represent certifications of taste and certifications of safety by individual third-party organizations.

In CA, we want to know the target respondents’ average preference for each aspect. Using the results of each respondent's selection of various options through a questionnaire survey, we can calculate the average β values for the different groups by the amount of information provided, type of drinking water currently consumed, socio-demographic factors, and conscious aspect. For the CA calculation, the conditional logit model was applied.

Design of the questionnaire

To learn about people's preferences regarding water, we used a questionnaire. The questionnaire consisted of five parts: (i) information provision, (ii) CA profiles, (iii) current drinking water, (iv) socio-demographics, and (v) conscious aspect.

Information provision

First, basic information about types of water, hardness, and certifications of taste and safety was provided to all the respondents in the questionnaire. The purpose of this information was to avoid misunderstanding of wordings and to share the same concepts for target attributes. An illustration with a one-sentence explanation was provided for each type of water. Four standard classes of hardness – namely ‘soft’, ‘middle-soft’, ‘hard’, and ‘extremely hard’ — were shown with their hardness ranges. The hardness values of popular bottled water and tap water in Tokyo, the USA, and Europe were also shown. For the certification categories, it was explained that water safety and taste were certified by a third party. In the case of safety, it was also explained that the water quality met the safety standard for each water type.

The respondents were divided into 12 groups and were given different additional information. Table 1 shows the combination of the information provided. The effect of these information provisions will be discussed using the models modified from the base model shown later.

Table 1

Grouping of respondents based on supplied information

GroupingGroup
123456789101112
Provided information Basic information* 
Tap water safety** – – – – – – 
Cost – – – – – – 
Cancer risk – – – – 
Profile design     
GroupingGroup
123456789101112
Provided information Basic information* 
Tap water safety** – – – – – – 
Cost – – – – – – 
Cancer risk – – – – 
Profile design     

X: provided. -: not provided, A: shown as a fraction. B: shown as a multiple of the WHO standard.

*Basic information about types of water, hardness, and certifications of taste and safety.

**Brief explanation about remaining chlorine and water quality standard.

For the information on tap water safety, it was explained that the remaining chlorine can prevent bacterial growth, that tests of tap water quality include fifty criteria that must be satisfied, and that tap water is strictly regulated by the Ministry of Health, Labour and Welfare in Japan.

For the cost information, the approximate annual running cost of 400 L was shown: 160 yen for tap water, 160 + 3,000 yen (cartridge) for filtered water, 40,000 yen for bottled water, and 40,000 + 12,000 yen (electric use) for water-dispenser water. As of July 2014, the yen–dollar exchange rate was 101.8 yen/USD.

For the cancer risk information, it was explained that the standard of drinking water is set at 10−5 by the World Health Organization (WHO) and that one excess cancer occurs in 100,000 people by drinking 10−5 cancer-risk water (consumed amount is involved) for seventy years.

After each item of information provided, a check box was inserted to confirm that a respondent had read the information.

CA profiles

In the second part of the questionnaire, to ask people's preferences regarding the water, pair-wise figures as shown in Figure 1 were used. Here, types of water, hardness, cancer risk, certification of taste, and certification of safety were selected as attributes (k) determining the total observable utility of drinking water (Vij). Two modes of expression for cancer risk were used: expression of the absolute value as a fraction (A) and as multiples of the WHO standard (B). In the case of (B), three levels were set: one-tenth of, equal to, and ten times the WHO standard. In the case of (A), it was shown as 1/1,000,000, 1/100,000, and 1/10,000, respectively. There were four set levels of hardness: soft, middle-soft, hard, and extremely hard. In the case of certification, there were two levels, ‘shown (1)’ and ‘not shown (0)’. The levels used for each attribute are summarized in Table 2. The combination of each attribute level was decided by orthogonal design using free software R and eight pair-wise questions were shown to each respondent.

Table 2

Attributes and level settings for CA profile design

Attribute (k)(xk) Levels
Types of water [TAP]*   tap water (1, 0), 
[FW]* filtered water (1, 0), 
[BT]* bottled water (1, 0), 
[WD]* water-dispenser water (1, 0) 
Hardness [HDN]  soft (1), middle-soft (2), hard (3), extremely hard (4) soft (1, 0), middle-soft (1, 0), hard (1, 0), extremely hard (1, 0)** 
Cancer risk [CNR] (A) 1/1,000,000 (1), 1/100,000 (2), 1/10,000 (3) 
(B) one-tenth of (1), equal to (2), and 10 times (3) the WHO standard 
Certification of taste [CT]  X (there is certification: 1), - (no certification: 0) 
Certification of safety [CS]  X (there is certification: 1), - (no certification: 0) 
Attribute (k)(xk) Levels
Types of water [TAP]*   tap water (1, 0), 
[FW]* filtered water (1, 0), 
[BT]* bottled water (1, 0), 
[WD]* water-dispenser water (1, 0) 
Hardness [HDN]  soft (1), middle-soft (2), hard (3), extremely hard (4) soft (1, 0), middle-soft (1, 0), hard (1, 0), extremely hard (1, 0)** 
Cancer risk [CNR] (A) 1/1,000,000 (1), 1/100,000 (2), 1/10,000 (3) 
(B) one-tenth of (1), equal to (2), and 10 times (3) the WHO standard 
Certification of taste [CT]  X (there is certification: 1), - (no certification: 0) 
Certification of safety [CS]  X (there is certification: 1), - (no certification: 0) 

*When analyzed, one of these terms is excluded to avoid multicollinearity.

**Use of dummy variable (1, 0) for each condition of hardness was also investigated.

Respondents were asked which water they would want to drink if they moved to new houses in the 23 wards of Tokyo. For the cancer risk description, the absolute value as a fraction was used in design A, while multiples of the WHO standard were used in design B (as shown in ‘Profile design’ in Table 1).

After they had answered these eight pair-wise questions, the respondents were also asked about which aspect they had mainly considered for the pair-wise choices: hardness, cancer risk, certification of taste, certification of safety, remaining chlorine, water quality standard, handling, and cost.

Current drinking water

In the third section of the questionnaire, respondents were asked which water they mainly drink at home now. They selected from tap water, filtered water, bottled water, and water-dispenser water.

Socio-demographics

In the final part of the questionnaire, questions were asked about socio-demographics factors, such as gender, age, family income, and parental status (existence of infants).

Conscious aspect

An analytic hierarchy process (AHP) method (Saaty 1994) was used to determine people's consciousness regarding each aspect of water. The hierarchical structure as shown in Figure 2 was considered, and respondents were first asked which of the following four criteria were more important in selecting water: taste, safety, cost, or handling. The aspect of handling was considered here because it is a qualitative aspect and could not be incorporated in the CA profile. The six (4C2 = 6: number of combinations selecting two from four options) pair-wise questions using the 5-item Likert's scale were asked of each respondent as shown in Figure 3. Furthermore, each respondent was asked which water he or she preferred from the viewpoint of one criterion (Figure 3). For each pair-wise result, the relative score ‘b’ (ranging from 0 to 1) for each aspect (e.g. b1: score for taste) among four aspects and the score ‘a’ (ranging from 0 to 1) for each water from the viewpoint of one aspect (e.g. a11: score of tap water from the viewpoint of taste) can be calculated. When those scores are multiplied and summed up for one water type (score of water m = : : l is criterion (1–4), m is water type (1–4)), the weight for the water type can be calculated.
Figure 2

Hierarchical structure considered for conscious aspects.

Figure 2

Hierarchical structure considered for conscious aspects.

Figure 3

Examples of AHP pair-wise questions. Q(a) shows comparison between the criteria, such as taste, safety, cost and handling. Q(b) shows comparison between the alternatives, such as tap water, filtered water, bottled water, and water-dispenser water, from the viewpoint of one of the criteria.

Figure 3

Examples of AHP pair-wise questions. Q(a) shows comparison between the criteria, such as taste, safety, cost and handling. Q(b) shows comparison between the alternatives, such as tap water, filtered water, bottled water, and water-dispenser water, from the viewpoint of one of the criteria.

Questionnaire survey

The questionnaire survey was conducted using an online questionnaire survey method, which has become popular in academic fields as well as in marketing. The designed questionnaire as explained above was sent to a survey company (Cross Marketing Co.) to make a web version of the questionnaire. Then they asked the people registered with the company to answer the questionnaire.

Prior to the main survey, a pilot-test using one group (G2: N = 800) was conducted during September 25 to 26, 2013 to check the wording and CA profiles. After the pilot-test, we conducted the main questionnaire survey answered by men and women aged 20 to 69 years living in the 23 wards of Tokyo. The number of respondents was set at 600 giving 4,800 samples (eight pair-wises for each respondent) for each group, corresponding to 7,200 respondents (for 12 groups) and 57,600 samples in total. The age and gender distributions were adjusted to coincide with those of the general population in Tokyo's 23 wards based on the National Census. We conducted the survey using Cross Marketing Co. from December 9 to 16, 2013.

RESULTS AND DISCUSSION

Base model

First, the base model using the basic equation as shown in Equation (2) was demonstrated. The resulting β values using this model are shown in the supplemental information (SI, Table S1, available with the online version of this paper). Each water type was evaluated relative to one excluded water type (shown as ‘-’). All groups except for G3 show that the image of tap water was the worst among the four water types. The negative values of βH indicate that people did not prefer the water with higher hardness. In Japan, the water hardness is relatively lower than in Europe (Takahashi & Imaizumi 1988). People usually drink soft or middle-soft water, which are also preferred for making tea and boiling rice in Japan. This can be an influence on people's negative preferences for high hardness. Larger negative values of βR indicate that people strongly avoided the water with a higher cancer risk. Besides, the larger positive values of βCT and βCS indicated that people attached high value to the certifications for safety and taste.

Among the G2 samples (N = 4,800), where all information was provided, the samples were divided into groups based on which aspects were considered for pair-wise choices, and β values of the base model were compared (see SI, Table S2, available online). As seen in each N, the samples who had focused on cancer risk (N = 1,528) and certification of safety (N = 904) reached more than half of the total samples. The β values coincided with the considered aspects; the samples who had focused on hardness (N = 768), cancer risk, and certification of safety showed larger absolute values of βH, βR, and βCS, respectively. The number of samples that considered remaining chlorine was small (N = 48); however, they evaluated the tap water utility as significantly lower. As with the cost information effects (θ*IC in Table 3), the people who were conscious of cost had a more negative assessment of bottled and water-dispenser water.

Table 3

Calculation results by the information provision models

 Calculated coefficients using Model I (Equation (3))
Calculated coefficients using Model II (Equation (4))
BaseAdded partsBaseAdded parts
βFW 0.47*** θFIS −0.04 0.42*** θFIR 0.004 
  θFIC 0.13    
βBT 0.46*** θBIS −0.06 0.40*** θBIR −0.72** 
  θBIC −0.66**    
βWD 0.64*** θWIS 0.06 0.13*** θWIR 1.17 
  θWIC −0.74***    
βH −0.17***   −0.39*** θHIR −0.27*** 
βR −0.51*** θR −0.22***    
βCT 0.43***   0.32*** θTIR −0.67*** 
βCS 0.35***   0.54*** θSIR −0.52*** 
Group G1, G2, G4, G5, G7, G8, G10, G11  G2, G3 
 38,400  9,600 
lnL  − 22,312  − 6,102 
ρ2  0.075  0.083 
 Calculated coefficients using Model I (Equation (3))
Calculated coefficients using Model II (Equation (4))
BaseAdded partsBaseAdded parts
βFW 0.47*** θFIS −0.04 0.42*** θFIR 0.004 
  θFIC 0.13    
βBT 0.46*** θBIS −0.06 0.40*** θBIR −0.72** 
  θBIC −0.66**    
βWD 0.64*** θWIS 0.06 0.13*** θWIR 1.17 
  θWIC −0.74***    
βH −0.17***   −0.39*** θHIR −0.27*** 
βR −0.51*** θR −0.22***    
βCT 0.43***   0.32*** θTIR −0.67*** 
βCS 0.35***   0.54*** θSIR −0.52*** 
Group G1, G2, G4, G5, G7, G8, G10, G11  G2, G3 
 38,400  9,600 
lnL  − 22,312  − 6,102 
ρ2  0.075  0.083 

**p < 0.05.

***p < 0.01.

Modification of the base model

To evaluate the effects of various factors, such as provided information, socio-demographics (without age) and conscious aspect, the results were recalculated using the models modified from the base model. In the base model, V is explained by the summation of partial utilities (β*x). In the modified model, it is assumed that people's preference (β) is changed like β (1 + α), which means that β can rise (1 + α)-fold by the factors. The α is described like Σ θ*y, where a dummy variable is used for y.

For the factors such as conscious aspect and age, grouping gave better results than the model modification; therefore, change of β using the base model in different groups is discussed.

Information provision effect

To evaluate the effects of the difference of provided information, two modified models were investigated. Model I is Equation (3). The bold parts are added items to Equation (2). 
formula
3

The intention of this modification was based on the assumption that the preference (β) on each attribute can be changed by ‘related’ information provision. Here, [IS] and [IC] are dummy variables for information provision about tap water safety and about the cost, respectively. When the information on safety or cost is provided, 1 is inserted in the term. For example, as seen in the first part, if the information on tap safety and water cost is provided ([IS] = 1, [IC] = 1), the preference for filtered water is changed from βFW to (1 + θFIS + θFICFW. Here, the additivity of information is assumed. [F/M] represents whether the cancer risk is provided as a fraction (1) or as a multiple of the WHO standard (0) and it is assumed that this influences just the preference regarding cancer risk (βR).

Model II modified from the base model is Equation (4). 
formula
4

The intention of this modification was based on the assumption that the preference regarding each attribute can be mitigated by ‘important’ information (cancer risk in this case). Here, [IR] is a dummy variable representing whether the cancer risk attribute exists (1) or not in the profile. The hypothesis of model II is that showing especially important information (such as cancer risk information) can dilute or increase the importance of other aspects.

The results are shown in Table 3. In Model I, no parameters about tap water safety information (θ*IS) were statistically significant. This means that the information about tap water safety did not have any influence on people's ideas about each water type. In the case of cost information, θBIC and θWIC were significantly negative. This indicates that cost information for bottled water and water-dispenser water, which tend to cost more than other water types, negatively influenced ideas about those types of water. It was found that θR was significantly negative. This indicates that the cancer risk expressed as a multiple of the WHO standard was perceived as a larger risk than the risk expressed as an absolute value using a fraction. Stone et al. (1994) focused on the low probability risk (tire blowout injury risk) and compared the incidence rate expression (e.g. ‘0.0000060’) and relative expression (e.g. ‘half that of standard tires’) formats. They determined that using a relative expression increased the perceived risk, while the absolute expression of a very small risk would be perceived as nil. It has also been reported that risk reduction expressed in a relative format is perceived as greater than when expressed in an absolute format (Visschers et al. 2009).

In Model II, θBIR, θHIR, θTIR, and θSIR showed significantly negative values. As seen in the base model, the negative perception of cancer risk (βR) is quite large. Therefore, it was indicated that the risk information is more important for consumers in selecting drinking water and, when the risk information is provided, the perceptions of other attributes can be diluted. A similar ‘dilution effect’ has also been reported in diagnostic information studies (Nisbett et al. 1981; Zukier 1982; Tetlock & Boettger 1989).

Current drinking water effect

To know the effect of current drinking water choices, the respondents were divided by their current drinking water and also divided into people who drink the water they want and those who drink a water other than the one they want, and their parameters were analyzed using the G2 group. As shown earlier, the trends of the G1 and G2 groups (groups to whom all information was provided) were not different; therefore we used one group where all information was provided (G2) for detailed analysis by incorporating several factors. The parameters were calculated as shown in Table S3 (available with the online version of this paper).

As expected, the preferences on current drinking water were higher than others. For example, the people who drink water-dispenser water show the highest preference for water-dispenser water (βWD = 0.733, p < 0.01).

Socio-demographic effects

Using the G2 samples, the effects of socio-demographic factors were investigated.

In the case of age, instead of a gradual effect, each age range shows a different preference independently; therefore, the coefficients were calculated for each age range using the base model, as shown in Equation (2). The results are shown in SI (Table S4, available online). In the case of water type, people in their fifties and sixties had more positive perceptions of filtered water. Some previous studies have shown that older people are more concerned about food risks than younger people (Lin 1998; Breakwell 2000; Dosman et al. 2001). Our results partially agreed with these previous studies, and the perceptions of people in their twenties of cancer risk information were much weaker than other generations, although the perceptions were quite similar among the different groups of people in their thirties and above.

To evaluate the influence of gender, income, and parental status (existence of infants) on water preference, Model III was applied (Equation (5)). 
formula
5

Here, [D] represents each socio-demographic variable: gender (male: 1, female: 0), income (1 to 8), and parental status (existence of infants, yes: 1, no: 0). For age and income evaluations, G2 (N = 4,800) samples were used. In the case of parental status, the respondents in their thirties involved in G2 and G5 (N = 1,216) were used.

No θ*D for income or parental status showed significant value. In the case of gender (see SI, Table S5, available online), θWD, θRD, θTD, and θSD showed negatively significant values. The negative θRD represents the lower perception of cancer risk for men. Gender effects on various risk perceptions have been well investigated, and many studies have revealed that women perceive food risks at a much higher rate than men (Lin 1998; Tetlock & Boettger 1989; Dosman et al. 2001). The negative θWD indicate that men had a more negative assessment of the utilities of water-dispenser water. This coincided with our previous study, where the proportion of women was larger in the clusters of people who prefer water-dispenser water (Amano et al. 2013). The negative θTD and θSD mean that men were less positively influenced by certifications from third parties. Blend & van Ravenswaay (1999) show that the probability of purchasing eco-labeled apples was significantly lower for men than for women. Miller & Unnevehr (2001) also show that women preferred pork that was certified safe more than men did.

Conscious aspect effect

Many studies have shown that personality traits, rather than socio-demographics, play the most important role in influencing people's attitudes; therefore, each respondent's conscious aspect was incorporated into the base model. Based on the AHP analysis, each respondent's evaluation of each water type from one viewpoint (among taste, safety, cost, and handling) was calculated and Model IV (Equation (6)) was proposed. 
formula
6

Here, [a b A] represents the estimated positive evaluation of water. For symbol a, T is tap water, F is filtered water, and W is water-dispenser water. For symbol b, T is taste, S is safety, C is cost, and H is handling. For example, [TTA] represents the evaluation of tap water from the viewpoint of taste. The hypothesis here is that personal evaluation of water from the viewpoint of b has an additive impact on each water type βa. The results are shown in SI, Table S6 (available online).

In the model, βa is negative; therefore, all negative θ**A values indicated that any kind of positive evaluation can decrease the negative image of the target water. In particular, a positive image of the taste can have a strong impact on the water image.

The relationship between [a b A] and βa θabA is shown in Figure 4. This shows the relationships between the current evaluation and degree of water utility change. As seen in TT, the current evaluation of tap water taste is lower; however, the change is large. This indicates that if people's evaluation of tap water taste can be improved, the selection of tap water can be considerably increased. A similar situation can be observed in the case of tap water safety (TS). If people can recognize that tap water is much safer, they will start selecting it. The degree of change is in the same range as the case of handling, but people already highly value the handling of tap water. The room for improvement is therefore limited in this case. In the consideration of cost, the lower cost of tap water is fully evaluated, and the degree of change is also lower. Therefore, advertising the cost aspect cannot be effective for tap water promotion.
Figure 4

Relationships between positive evaluation and total image change. [a b A] represents the estimated positive evaluation of water [a] from the viewpoint of [b]. βaθabA represents the change rate of the positive evaluation. TT: tap water taste, TS: tap water safety, TC: tap water cost, TH: tap water handling. FT: filtered water taste, FS: filtered water safety, FC: filtered water cost, FH: filtered water handling. WT: water-dispenser water taste, WS: water-dispenser water safety, WC: water-dispenser water cost, WH: water-dispenser water handling.

Figure 4

Relationships between positive evaluation and total image change. [a b A] represents the estimated positive evaluation of water [a] from the viewpoint of [b]. βaθabA represents the change rate of the positive evaluation. TT: tap water taste, TS: tap water safety, TC: tap water cost, TH: tap water handling. FT: filtered water taste, FS: filtered water safety, FC: filtered water cost, FH: filtered water handling. WT: water-dispenser water taste, WS: water-dispenser water safety, WC: water-dispenser water cost, WH: water-dispenser water handling.

CONCLUSIONS

The influence of various attributes and information provision effects on drinking water selection were evaluated using CA. Among the evaluated attributes, risk and certification attributes had larger impacts on drinking water selection.

In terms of risk expression, the cancer risk of drinking water is categorized into low probability risk and relative expression as shown in multiples of the standard value showed more impact on consumers’ perceptions of cancer risk. Therefore, when we deal with relatively lower risk, as shown in this paper, we need to pay attention to the way the values are expressed.

Among socio-demographics, although gender has significant effects and age also has some influence on drinking water selection, other socio-demographic factors did not have significant influences. On the other hand, the personal conscious aspect for drinking water had more significant impacts on drinking water selection. Incorporating information and AHP effects in the CA framework can give helpful insights into which attributes or information should be considered or focused on for advertising tap water. Instead of using direct questions, using the CA framework can give a quantitative evaluation of each attribute based on the people's selection of the target goods or service. Besides, as we demonstrated, the information provision effect can be evaluated by using the CA framework. Policy makers, as well as water suppliers, can use our results and methodologies to improve communication with citizens about safety and selection of tap water as drinking water.

ACKNOWLEDGEMENTS

We would like to express our appreciation for the Asahi Glass Foundation's financial support for this study.

REFERENCES

REFERENCES
Alriksson
S.
Öberg
T.
2008
Conjoint analysis for environmental evaluation–a review of methods and applications
.
Environmental Science and Pollution Research
15
(
3
),
244
257
.
Amano
I.
Kurisu
K.
Hanaki
K.
2013
Classification of citizens in Tokyo by choice of drinking water and the reason
.
Journal of Japan Society of Civil Engineers, Ser. G (Environmental Research)
69
(
6
),
II_117
II_126
.
Baker
G. A.
Burnham
T. A.
2001
Consumer response to genetically modified foods: market segment analysis and implications for producers and policy makers
.
Journal of Agricultural and Resource Economics
26
(
2
),
387
403
.
Blend
J. R.
van Ravenswaay
E. O.
1999
Measuring consumer demand for ecolabeled apples
.
American Journal of Agricultural Economics
81
(
5
),
1072
1077
.
Breakwell
G. M.
2000
Risk communication: factors affecting impact
.
British Medical Bulletin
56
(
1
),
110
120
.
de França Doria
M.
Pidgeon
N.
Hunter
P. R.
2009
Perceptions of drinking water quality and risk and its effect on behaviour: a cross-national study
.
Science of the Total Environment
407
,
5455
5464
.
Dosman
D. M.
Adamowicz
W. L.
Hrudey
S. E.
2001
Socioeconomic determinants of health- and food safety-related risk perceptions
.
Risk Analysis
21
(
2
),
307
317
.
Grubb
E. L.
Hupp
G.
1968
Perception of self, generalized stereotypes, and brand selection
.
Journal of Marketing Research
V
,
58
63
.
Jacoby
J.
Olson
J. C.
Haddock
R. A.
1971
Price, brand name, and product composition characteristics as determinants of perceived quality
.
Journal of Applied Psychology
55
(
6
),
570
579
.
Lee
M. K. O.
Turban
E.
2001
A trust model for consumer internet shopping
.
International Journal of Electronic Commerce
6
(
1
),
75
91
.
Lin
C.-T. J.
1998
Demographic and socioeconomic influences on the importance of food safety in food shopping
.
Agricultural and Resource Economics Review
24
(
2
),
190
198
.
Miller
G. Y.
Unnevehr
L. J.
2001
Characteristics of consumers demanding and their willingness to pay for certified safer pork
.
Journal of Agribusiness
19
(
2
),
101
119
.
Mitsukan Co.
2013
Survey about life consciousness on water
. ).
Saaty
T. L.
1994
Fundamentals of decision making and priority theory with the analytic hierarchy process Vol. VI of the AHP series
.
RWS Publications
,
Pittsburgh
,
USA
.
Stone
E. R.
Yates
J. F.
Parker
A. M.
1994
Risk communication: absolute versus relative expressions of low probability risks
.
Organizational Behavior and Human Decision Processes
60
,
387
408
.
Suntory Co
.
2013
(
accessed 25 July 2014)
.
Takahashi
Y.
Imaizumi
Y.
1988
Hardness in drinking water
.
Eisei Kagaku
34
(
5
),
475
479
.
TBW (Tokyo Metropolitan Government Bureau of Waterworks)
2013
Survey on customer satisfaction with water supply. http://www.waterworks.metro.tokyo.jp/customer/jigyo/h25mzoku/51.pdf
(
accessed 25 July 2014)
.
Tetlock
P. E.
Boettger
R.
1989
Accountability: a social magnifier of the dilution effect
.
Journal of Personality and Social Psychology
57
(
3
),
388
398
.
Visschers
V. H. M.
Meertens
R. M.
Passchier
W. W. F.
de Vries
N. N. K.
2009
Probability information in risk communication: a review of the research literature
.
Risk Analysis
29
(
2
),
267
287
.

Supplementary data