A. Subjects

The subjects who participated in water tasting were all undergraduate students at The American University of Paris, a small international university in Paris, France. They were drawn from three general education courses taught by the authors of this article (specifically, an environmental science course in which the water tasting was a lab for their class, as well as principles of micro- and macro-economics courses for which they received extra credit if they participated). They did not receive monetary compensation, they were warned of the risks of drinking excessive amounts of water, and they all signed consent forms, as required by the human subjects research committees that approved our research.

The students' responses to preliminary demographic questions are summarized in Table 1. Our subjects are not necessarily representative of larger populations, but as shown in that table, they were diverse in terms of their nationalities, which is consistent with the university overall. For an open-response question about their nationality, 39% identified at least one of their nationalities as the United States, 16% said the same about France, 4% identified as both French and American, and the rest (49%) identified one or more other nationalities that ranged from Albanian to Zimbabwean. Chinese were the third-largest group at 5%. The participants were young and mostly female, which is again consistent with the university overall. We did not ask about socioeconomic status, but it can be assumed that most are of high status because the university's tuition is high (about €30,000 per academic year) and financial aid is limited.

Table 1 Participant Information

Note: This table shows preliminary demographic information about the subjects who participated in our study.

Water is essential to life, so our subjects must consume water on a regular basis. They should nevertheless be considered novice rather than expert water tasters. Indeed, as shown in Table 1, less than 20% of our subjects thought their ability to distinguish different bottled waters would be above average. They were more confident about their ability to distinguish different wines, and even more confident about their ability to distinguish tap from bottled water. Those self-reported abilities are positively correlated, which will be shown as part of Table 5.

B. Training

Given that our subjects should be considered novice water tasters, we gave them a brief training in water tasting before our blind tasting. The goal of the training was to dispel any pre-existing belief among participants that all waters taste the same while attuning their palates to the tastes of water. The Berkeley Springs tasting also trains its judges, some of whom are novice tasters (Fulcher, Reference Fulcher2017; Lewis-Kraus, Reference Lewis-Kraus2006). Our training was based on a YouTube video of a non-blind tasting the water sommelier Riese organized for a Zagat reporter (Zagat, 2015). In the video, Riese serves the reporter three brands of bottled water with different levels of carbonation and different amounts of “Total Dissolved Solids” (TDS), which include “carbonates, bicarbonates, chlorides, sulfates, phosphates, nitrates, calcium, magnesium, sodium, potassium, iron, manganese, and a few other minerals” (Mascha, Reference Mascha2006, pp. 35–37). He starts by serving the still version of the VOSS brand of bottled water, which has a relatively low TDS level of 44 milligrams per liter (mg/L). The reporter says the water just tastes like water. Next, Riese serves the still version of the Iskilde brand with a higher TDS level of 426 mg/L. The reporter reacts by saying that, to her surprise, she can actually taste the difference. The third water he serves is the sparkling Vichy Catalan brand with an extremely high TDS level of 3,052 mg/L. She reacts strongly, saying it tastes like Alka-Seltzer. Of note, all three of those bottled waters are included on the water menu at Ray's and Stark Bar, and those waters cover almost the entire range of TDS levels on that menu. Among the waters on the menu, VOSS has the second lowest TDS level, Iskilde has the highest TDS level among the still waters, and Vichy Catalan has the highest TDS level.

The participants in our study were shown that video and served the same three brands of bottled water in a non-blind fashion as they followed along. Anecdotally, their reactions were similar to the reporter's. Some subjects said they could not taste a difference between the VOSS and Iskilde brands, but many said they could, and everyone said the Vichy Catalan brand was noticeably different. The reporter's reactions could have influenced our subjects', but it seems safe to conclude that waters with high enough amounts of carbonation and dissolved minerals can be distinguished from waters with low amounts. Whether subtler differences can be discerned is a question addressed by our study.

Immediately after the non-blind training, a blind taste test was conducted. None of the waters used in the training were used in the blind tasting, and participants were told that beforehand. The brands of bottled water used in the blind tasting were masked from participants throughout and only unmasked afterwards.

C. Waters for Blind Tasting

There are thousands of brands of bottled water (Mascha, Reference Mascha2006, p. 20), but only a few can be served during any tasting. We selected four based on the following criteria. First, to ensure a selection of distinct, high-quality waters, we only considered brands included in both Ray's and Stark Bar's water menu and Mascha's (Reference Mascha2006) guide to fine water. Next, we considered only still waters because brands of sparkling water can vary in their level of carbonation in a way that could perhaps be detected visually rather than by taste. After excluding the still waters used in our non-blind training, we picked waters that varied widely in their TDS levels. The water sommelier at Ray's and Stark Bar (see, e.g., Zagat, 2015) and Mascha (Reference Mascha2006, p. 35) both emphasize that a water's TDS is the most important dimension to its taste. Finally, we chose waters available at La Grande Épicerie de Paris, an upscale grocery store in Paris known for bottled water selection.

Based on those criteria, we selected the Speyside Glenlivet, Acqua Panna, Fiji, and Hildon brands, which vary in their TDS levels from 58 to 312 mg/L. Additional information about those bottled waters—including their sodium, calcium, magnesium, and hardness, where a water's “hardness” is a weighted sum of the amount of calcium and magnesium in the water (see, e.g., Mascha, Reference Mascha2006, p. 38)—is given in Table 2. That information is taken from the water menu at Ray's and Stark Bar.

Table 2 Water Information

Note: This table shows information about the waters included in our blind taste test. The information on the bottled waters is taken from Ray's and Stark Bar (2015), while the information on the tap water is taken from the public water supply company Eau de Paris. The mineral contents are all measured in milligrams per liter.

Tap water was also used for comparison in some of the experiments discussed later. As with bottled water, there are numerous tap waters. We used the tap water that was the easiest to obtain: the water that could be drawn from a sink in the building in which we conducted our experiments. Additional information about the tap water is given in Table 2; that information is from the public water supply company for the city of Paris.

D. Other General Procedures

The blind tasting, as well as the non-blind training discussed earlier, was conducted in a science teaching lab. Due to the small size of the lab, experiments were run several times in the Spring and Fall 2016 semesters with groups ranging from five to 20. It was not feasible to isolate subjects in that lab, so we did the following to limit their influence on one another. Before subjects arrived, we set up all the experiments. Each subject's setup looked identical, but the waters inside each glass were randomized with the aid of a computer program. We informed everyone of that randomization, emphasizing that different waters would be in different glasses for each subject. Therefore, any opinions about a glass of water expressed by one person should not influence the opinions of others. We kept track of the setups by assigning an identification number to each one. The large number of random setups meant that we—the researchers—were unable to remember during the tasting how the waters were set up. The test was therefore “double blind” in the sense that neither the researchers nor subjects knew which waters were which.

Between the non-blind training and blind tasting, each subject was given a total of 20 glasses of water (specifically, 3 glasses for the training, 12 for the first experiment discussed later, and 5 for the other experiments). The Berkeley Springs tasting asks its judges—which again include trained but still novice tasters—to evaluate 20 glasses of water at the same time and to do so four times in one day (Fulcher, Reference Fulcher2017; Lewis-Kraus, Reference Lewis-Kraus2006, p. 26). Thus, the number of glasses we served our subjects is well within the range of an official tasting event. Neutral water crackers were available as palate cleaners. The Berkley Springs tasting also makes water crackers available to its tasters (Lewis-Kraus, Reference Lewis-Kraus2006, p. 26).

According to Mascha (Reference Mascha2006, p. 56), the recommended temperature for serving still water is 12°C. Maintaining that cold temperature for all the waters throughout the experiment was infeasible, so we served all of the waters at room temperature. The Berkeley Springs tasting also serves still waters at room temperature (Fulcher, Reference Fulcher2017).

A. Method

For the first experiment, we followed similar studies of wines (Ashton, Reference Ashton2014; Weil, Reference Weil2001, Reference Weil2005, Reference Weil2007), beers (Almenberg, Dreber, and Goldstein, Reference Almenberg, Dreber and Goldstein2014), and waters (Dietrich and Gallagher, Reference Dietrich and Gallagher2013; Gallagher and Dietrich, Reference Gallagher and Dietrich2010) by using so-called “triangle tests” to assess whether participants could tell the difference among bottled waters. In a triangle test, subjects were given three small glasses of water that looked identical. Two of the glasses had the same brand of bottled water in them (the “twin”). The third glass had a different brand of bottled water (the “singleton”). Participants were asked to carefully taste the waters to try to identify the singleton. We emphasized there was no deception involved; one water was different from the other two, even if the differences might be subtle. A subject has a one-in-three chance of correctly identifying the singleton even if they cannot tell a difference.

We gave each subject four sets of triangle tests laid out on a clearly marked template.Footnote ¹ They were instructed to move sequentially from one triangle test to the next, trying to identify the singleton each time. Each triangle test had a different pair of bottled waters. In the video mentioned earlier (Zagat, 2015) and a similar video (Guardian Food, 2015), Riese moves from waters with lower to higher TDS levels when conducting a tasting. The triangle tests were therefore arranged by their TDS levels from lowest to highest, except for the last triangle test, which involved comparing the waters with the lowest and highest TDS levels. Although participants were not informed of it, for every triangle test except the last, the singleton was the water with the relatively higher TDS level.

In addition to asking our subjects to identify the singletons, we asked them two questions for each triangle test. First, we asked whether they were confident they correctly identified the singleton. They ranked their confidence on a five-item Likert-type scale ranging from “no confidence at all” to “completely confident.” Second, we asked which water they would prefer—the water they identified as the twin or the one they identified as the singleton—as their everyday drinking water. They expressed their preference on a five-item scale with “indifferent between them” as the middle option. We prompted them to consider their preferences in terms of an “everyday drinking water” to make the results more comparable to the second experiment we conducted.

B. Results and Discussion

The results of our triangle tests are shown in Table 3. The participants did better than random chance for each test, and statistically significantly so for every test, except the first one, which involved the two waters with the lowest TDS levels.

Table 3 Results of Triangle Tests

Note: This table reports the results of four triangle tests between different bottled waters. Participants should correctly identify the singleton one-third of the time by chance alone, so we indicate if the percentage who identified the singleton is statistically significantly greater than that based on the adjusted Chi-squared test suggested by Heymann and Lawless (Reference Heymann and Lawless1999, pp. 131–132).

* p < 0.10, ** p < 0.05, *** p < 0.01.

Table 4 shows the percentage of singletons participants should correctly identify by random chance and the percentage they correctly identified. More participants correctly identified all four singletons than would be expected by chance, and fewer failed to identify any than would be expected. Also, more participants correctly identified two or three than would be expected by chance, and fewer identified only one.

Table 4 Frequency of Identified Singletons

Note: This table shows, for all four triangle tests, the percentage of subjects who would be expected by chance alone to correctly identify a given number of singletons, the percentage who actually did, and whether the difference between those percentages is statistically significant based on a Chi-squared test. The percentages were rounded.

* p < 0.10, ** p < 0.05, *** p < 0.01.

Although our subjects were better than chance at identifying the singletons, the singleton was correctly identified less than half of the time for each triangle test, including the last test with the biggest TDS difference (see Table 3). Moreover, our subjects correctly identified the singleton in less than half of the triangle tests—about 1.8 out of the four—on average (see Table 4). Those results suggest our subjects were only slightly better than chance at distinguishing bottled waters.

Similar results have been found for wine and beer tastings. Using triangle tests with various wines and mostly non-expert tasters, Ashton (Reference Ashton2014) and Weil (Reference Weil2001, Reference Weil2005, Reference Weil2007) found the singleton was usually identified no more than around half of the time. Almenberg, Dreber, and Goldstein (Reference Almenberg, Dreber and Goldstein2014) also found similar results for three triangle tests with pale European lagers. Their subjects were not statistically significantly better than chance at distinguishing either Heineken or Czechvar beer from Stella Artois, and they actually did worse than would be expected by chance, although not significantly so. For the one test in which they did statistically significantly better than chance, they correctly distinguished Heineken from Czechvar beer only 48% of the time (Almenberg, Dreber, and Goldstein, Reference Almenberg, Dreber and Goldstein2014, pp. 4–5).

Our results can also be compared to previous triangle tests of water. Gallagher and Dietrich (Reference Gallagher and Dietrich2010) used three undisclosed brands of bottled water with TDS levels of 3, 31, and 524 mg/L, respectively. Out of three triangle tests they conducted with those waters at room temperature and untrained tasters, the singleton was correctly identified as frequently as 66% of the time (when comparing the 31 and 524 mg/L waters) and infrequently as 36% of the time (when comparing the 3 and 31 mg/L waters). Across all three tests, 50% of their subjects correctly identified the singleton (Dietrich and Gallagher, Reference Dietrich and Gallagher2013, p. 22).

In another study, Dietrich and Gallagher (Reference Dietrich and Gallagher2013) took an undisclosed brand of bottled water with a TDS level of 524 mg/L and diluted it with filtered water to obtain waters with various TDS levels. Out of nine triangle tests they conducted with those waters at room temperature and untrained tasters, the singleton was correctly identified as frequently as 65% of the time when comparing the undiluted bottled water to a diluted water with a TDS level of 31 mg/L. The singleton was identified as infrequently as 39% of the time when comparing waters with a smaller, although not the smallest, TDS difference. Across all nine tests, 50% of their subjects correctly identified the singleton (Dietrich and Gallagher, Reference Dietrich and Gallagher2013, p. 259).

It is noteworthy that Dietrich and Gallagher (Reference Dietrich and Gallagher2013) found a close but not monotonic relationship between the difference in TDS levels and the percentage of subjects who correctly identified the singleton. They found that result even though they were diluting the same brand of bottled water (rather than comparing different brands) and even though different subjects (rather than the same ones who might get fatigued) participated in each triangle test. The fact that subjects in our study did not perform the best on the last triangle test with the highest TDS difference is therefore not necessarily a sign of fatigue. Even if it is a sign of fatigue, that raises questions about the ability of Berkeley Springs judges to evaluate upwards of 80 glasses of water in one day.

Thus, our subjects performed about as well as those in previous triangle tests of wines, beers, and waters. Some of our subjects performed better than others, yet those observed abilities had either no correlation or a slight negative correlation with their self-report abilities. Table 5 shows the rank correlation between each subject's self-reported ability to distinguish different wines or waters and the number of singletons they correctly identified in the triangle tests. Curiously, there is a small but statistically significantly negative correlation between their self-reported ability to distinguish bottled waters and their observed ability (ρ _s = −0.19, p-value = 0.05). Less confident subjects were perhaps more careful during their tasting.

Table 5 Correlations between Self-Reported and Observed Tasting Abilities

Note: This table shows the Spearman rank-correlation coefficient for 113 subjects' self-reported ability to distinguish between different wines, tap and bottled water, and different bottled waters, as well as their observed ability to distinguish different bottled waters based on the number of singletons they correctly identified in the triangle tests. Five subjects did not respond to one or more of the preliminary questions about their tasting abilities.

* p < 0.10, ** p < 0.05, *** p < 0.01.

As previously noted, in addition to asking subjects to try to identify the singleton in a triangle test, we asked if they were confident they correctly identified the singleton and whether they preferred the singleton they identified or the twin. Part of Table 3 shows, for each triangle test, the preferences expressed by subjects who correctly identified the singleton and expressed more than no confidence in their choice. We restrict our attention to those confident discerners because preferences expressed by subjects who misidentified the singleton or randomly guessed correctly are less interesting.

Many of the confident discerners expressed little or no preference between the bottled waters they tasted. At least 20% and as many as 33% said they were indifferent between the waters in each triangle test. No more than 16% expressed a “strong” preference over any of the waters. That said, among those who expressed any preference, some waters were more popular than others. In addition to most of the confident discerners preferring Acqua Panna to Fiji, most preferred Fiji to Hildon, and most also preferred Hildon to Speyside Glenlivet in their respective triangle tests. Similar preferences were expressed in the next experiment we conducted, which involved comparing all the waters on a simultaneous, rather than pairwise, basis.

A. Method

The second experiment was conducted immediately after the first one with the same participants. For that experiment, we gave each subject five glasses of water on another clearly marked template. Four of the glasses contained the four brands of bottled water used in the triangle tests. A fifth glass contained tap water. We informed our subjects that each glass had a different water, but did not inform them that one water was tap and the others were the same bottled waters from the triangle tests. As before, the glasses of water looked identical, they were laid out ahead of time and kept at the same temperature, the position of each type of water was randomized, and subjects were informed of that randomization.

The subjects were asked to rate each of those five waters on the 14-point scale used as part of the Berkeley Springs tasting. On that scale, a water should be assigned the highest rating of 14 points if you can say, “This water tastes really good. I would be very happy to have it for my everyday drinking water.” A water should be assigned the lowest rating of 1 point if it makes you say, “This water has a terrible, strong taste. I can't stand it in my mouth.” The full scale can be found in Lewis-Kraus (Reference Lewis-Kraus2006).

We also asked participants to use their ratings to rank the waters from their most to least preferred. They were strongly encouraged to break any ties if they gave the same rating to more than one water.

Asking our subjects to evaluate five glasses of water at the same time is somewhat demanding, but again the Berkeley Springs tasting asks its judges to simultaneously evaluate 20 glasses of water. Olkin et al. (Reference Olkin, Lou, Stokes and Cao2015, p. 24) also suggest that, when it comes to wine, five is a small enough number for most people to simultaneously evaluate.

B. Results and Discussion

Table 6 shows the distribution of rankings assigned to the five waters. Specifically, it shows the percentage of subjects who ranked a water as their most preferred, their second-most preferred, and so forth. That table reflects the rankings of only 110 of our 118 subjects. We ignored the rankings by eight participants who gave the same ranking to two or more waters because those ties may reflect “lazy tasting” (to borrow a phrase from Quandt, Reference Quandt2006, p. 9).

Table 6 Distribution of Rankings

Note: Our subjects ranked five different waters. This table shows the percentage of subjects who gave a water a given ranking, as well as the sum of the rankings for each water. Eight sets of rankings with one or more ties were ignored. The rank-sum for the tap water is about 1.1 times higher than the average rank-sum of the bottled waters, which is statistically significantly higher based on Quandt's (Reference Quandt2007) rank-sum test (p-value ≈ 0.03 based on 10,000 Monte Carlo simulations).

The Fiji brand of bottled water was given the most first-place rankings with a quarter of participants ranking it as their most preferred. Tap water was given the most last-place rankings with 29% ranking it as their least preferred. However, a quarter of participants also said that Fiji was their least preferred, and almost 20% said that tap water was their most preferred. Thus, there is no clear consensus about which waters are preferable to others.

One way to try to find consensus is by examining the rank-sums for each water (i.e., the sum of the ranks that subjects assigned to each water) shown in Table 6. If a subject assigns a ranking lower than first place, then that can be thought of as a vote against the water, as discussed by Quandt (Reference Quandt2006). The water that received the fewest votes against it—and was the most preferred in that sense—was Acqua Panna. The waters that received the most votes against them were Speyside Glenlivet and tap, which were the waters with the lowest and highest TDS levels, respectively.

In terms of the ratings underlying the rankings, histograms of the ratings given to each of the five waters are shown in Figure 1. The subjects made full use of the scale with ratings as low as 1 and as high as 14, but the average ratings were similar. The average (standard deviation) of the ratings assigned to Acqua Panna, Hildon, Fiji, Speyside Glenlivet, and tap water were 10.5 (2.5), 10.2 (2.6), 9.8 (3.2), 9.6 (2.8), and 8.9 (3.4), respectively. Note the bottled water with the highest average rating, Acqua Panna, was rated less than 1 point higher on the 14-point scale than the bottled water with the lowest average rating, Speyside Glenlivet.

Note: This figure shows, for each water subjects rated on the 14-point scale, a histogram of the ratings given to those waters. The dashed line indicates the mean rating.

Figure 1 Distribution of Ratings

To try to understand why some waters were preferred to others, we ran the simple ordinary least squares (OLS) reported in Table 7. The dependent variable in each regression is the rating a subject gave a water on the 14-point scale. Almenberg and Dreber (Reference Almenberg and Dreber2011), Ashenfelter and Jones (Reference Ashenfelter and Jones2013), and Goldstein et al. (Reference Goldstein, Almenberg, Dreber, Emerson, Herschkowitsch and Katz2008) ran similar OLS regressions for wine ratings. Fixed effects for each subject were used in case some subjects tended to rate all the waters higher or lower on the 14-point scale. Dummy variables for the position of a water were also included to see whether the way in which the waters were arranged—from left to right in a row—affected their rating. None of the positional dummies were statistically significant (either individually or jointly) in any of our regressions, which suggests the arrangement of the waters did not bias the ratings.

Table 7 OLS Results for Ratings

Note: This table reports the results of OLS regressions for the rating a subject gave a water on the 14-point scale. Fixed effects were used for each subject, and standard errors were clustered by subject. The number of observations corresponds to 118 subjects, all of whom rated each water. The four positional dummies are not jointly significant in the first (F = 0.83, p-value = 0.50), second (F = 0.65, p-value = 0.63), third (F = 0.63, p-value = 0.64), or fourth (F = 0.91, p-value = 0.46) regression.

* p < 0.10, ** p < 0.05, *** p < 0.01.

For the first regression in Table 7, dummy variables for the different waters were included. The omitted dummy was tap water. We see that Acqua Panna and Hildon tended to be rated more than 1 point higher than tap water on the 14-point scale, Fiji was rated somewhat higher than tap, and Speyside Glenlivet was rated similarly to tap. We also see again that Acqua Panna tended to be rated less than 1 point higher than Speyside Glenlivet, even after controlling for subject-specific fixed effects.

The waters differ in terms of their TDS levels, so the second regression in Table 7 includes both the TDS level of a water and the square of its TDS level to capture a nonlinearity. The regression implies that, other things being equal, a water with TDS level of about 210 mg/L would be rated the highest. Waters with higher or lower TDS levels would be rated lower. That finding is consistent with Teillet et al. (Reference Teillet, Urbano, Cordelle and Schlich2010) who suggest waters with intermediate levels of mineralization are the most preferred.

The third regression in Table 7 includes some of the minerals that make up a water's TDS level. Specifically, the regression includes the water's sodium level and hardness. The regression implies that (again, other things being equal) a water with a sodium level of about 13 mg/L would be rated the lowest. A water with a hardness of about 182 mg/L would be rated the highest.

The price of a water can be included in a regression like the ones in Table 7, although the price of tap water will obviously be an outlier. The prices we paid at La Grande Épicerie de Paris for the Acqua Panna, Hildon, Speyside Glenlivet, and Fiji bottled waters were about 2.93, 3.00, 3.07, and 3.75 euros per liter, respectively. By comparison, the tap water was free to us and, although someone was ultimately paying for it, the cost of tap water in France has been estimated to be a mere 0.0038 euros per liter (Brei, Reference Brei2018, p. 2).

The final regression in Table 7 follows Goldstein et al. (Reference Goldstein, Almenberg, Dreber, Emerson, Herschkowitsch and Katz2008) by including the log of the price of the waters in euros per liter. Goldstein et al. (Reference Goldstein, Almenberg, Dreber, Emerson, Herschkowitsch and Katz2008) found that, in double-blind tastings with a variety of wines, non-experts tended to rate cheaper wines as better, even after controlling for individual fixed effects. We find that more expensive waters were rated as better, although the statistically significant effect is arguably not substantively significant. Our regression implies the most expensive bottled water we served our subjects would be rated only about 1.1 points higher on the 14-point scale than the inexpensive tap water.

Moreover, if we re-ran the last regression in Table 7 with only bottled waters, then the estimated coefficient on the log of the price of a bottled water would be about −1.50 (s.e. = 1.73). That estimate would not be significantly different than zero (p-value = 0.39), but it suggests there is no correlation or perhaps even a negative correlation between the price of a bottled water and its rating.

A. Method

In our final experiment, we gave participants five written descriptions corresponding to each of the five waters they had rated in the previous experiment. We asked them to try to match each water to its corresponding description. Note that a randomly guessing subject would have a one-in-five chance of correctly matching a given water to its description.

The description we gave participants for the tap water was simply that it was tap water. The descriptions of the bottled waters, which were taken from the water menu at Ray's and Stark Bar, are reproduced as part of Table 8. Those are expert descriptions in the sense that they were curated by the water sommelier who designed that menu.

Table 8 Results of Description-Matching Test

Note: This table shows the percentage of subjects who correctly matched a water to its description and whether that percentage is statistically significantly greater than 20% based on a Chi-square test. The descriptions of the bottled waters were taken from Ray's and Stark Bar (2015), except brand names and geographic locations that could reveal the brands were redacted from the descriptions given to our subjects.

* p < 0.10, ** p < 0.05, *** p < 0.01.

B. Results and Discussion

The results of that experiment are reported in Table 8. For each water, the participants were not significantly better than chance at matching the water to its description, except for Acqua Panna, but even in that case, less than 30% of participants were able to correctly match it to its description.

The fact that only 24% of participants correctly identified the tap water also means that 76% mistook a bottled water for tap water. Speyside Glenlivet, Hildon, Acqua Panna, and Fiji were mistaken for tap 28%, 27%, 24%, and 18% of the time, respectively. Those percentages are all close enough to 25% that we should perhaps not overanalyze them, but it is interesting that the two bottled waters most likely to be mistaken for tap were very different in their TDS levels yet similar in that they received lower ratings on average. Bad water is tap water, our subjects might have presumed.

As further evidence that participants were no better than chance at matching the descriptions, Table 9 shows the percentage of descriptions that participants would be expected to correctly match by random chance and the percentage they correctly matched. Those percentages are not significantly different.

Table 9 Frequency of Matched Descriptions

Note: This table shows the percentage of subjects who would be expected by chance alone to correctly match a given number of waters to their description and the percentage who actually did. The expected percentages were calculated as in the Montmort matching problem. The differences between those percentages are neither jointly nor individually significant at the 10% level based on Chi-square tests.

The expert descriptions of the bottled waters given in Table 8 are admittedly not very descriptive of their tastes. More detailed descriptions, like the ones given to wines, could perhaps be easier to match. Yet previous studies suggest that wine tasters, especially novice ones, are no better than chance at matching wines to descriptions (Storchmann, Reference Storchmann2012, pp. 24–25). Better descriptions would not necessarily improve the performance of our subjects, whose performance is consistent with previous findings for wine.