Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Food Quality and Preference 21 (2010) 967–976

Contents lists available at ScienceDirect

Food Quality and Preference


journal homepage: www.elsevier.com/locate/foodqual

Sensory methodologies and the taste of water


Eric Teillet a,c,*, Pascal Schlich b, Christine Urbano a, Sylvie Cordelle a, Elisabeth Guichard d
a
CNRS, UMR 5171, Centre Européen des Sciences du Goût, CNRS-UB-INRA, France
b
INRA, UMR 5171, Centre Européen des Sciences du Goût, CNRS-UB-INRA, France
c
Lyonnaise des Eaux, France
d
INRA, UMR 1129, FLAVIC, INRA-ENESAD-UB, France

a r t i c l e i n f o a b s t r a c t

Article history: Describing the taste of water is a challenge since drinking water is supposed to have almost no taste. In
Received 23 August 2009 this study, different classical sensory methodologies have been applied in order to assess sensory char-
Received in revised form 29 April 2010 acteristics of water and have been compared: sensory profiling, Temporal Dominance of Sensations
Accepted 29 April 2010
and free sorting task. These methodologies present drawbacks: sensory profile and TDS do not provide
Available online 7 May 2010
an effective discrimination of the taste of water and the free sorting task is efficient but does not enable
data aggregation. A new methodology based on comparison with a set of references and named ‘‘Polar-
Keywords:
ized Sensory Positioning” (PSP) has been developed enabling to easily define the sensory characteristics
Polarized Sensory Positioning
Drinking water
of water without presenting too many samples. Finally, this method provides a new type of sensory data
Sensory analysis requesting dedicated data analysis.
Ó 2010 Elsevier Ltd. All rights reserved.

1. Introduction odologies have been then adapted to water samples by Krasner,


McGuire, and Ferguson (1985) and Suffet et al. (1988) in a standard
Objectively determining the taste of water is not an easy task, method (AWWA, 1993). This methodology uses the ‘‘drinking
since drinking water by definition is supposed to have no particu- waters taste and odor wheel” (see for example Suffet, Khiari, & Bru-
lar taste. Indeed, the World Health Organization (WHO) has issued chet, 1999) which is a thorough list of sensory descriptors encoun-
recommendations to define criteria of comfort and pleasure (water tered in the taste of water. Despite efficient in identifying the off-
pleasant to drink, clear and with a balanced mineral content). flavors of waters due to pollution episodes, the high number of
These recommendations are the basis used by the European Union descriptors and their nature (mostly odors) make this method less
to prepare directives (JOCE, 1998) to define drinking water as lack- adapted to the determination of the taste of waters free of off-
ing any particular taste. flavors.
It is nevertheless clear that beside ‘‘off-flavors” produced by In a project aiming at understanding preference and behavior
pollution episodes, a variety of tastes can be detected in tap and towards drinking water, it was necessary to first understand the
bottled waters from various origins. Early, Bruvold and Pangborn sensory space of drinking water. In this context, several ‘‘classical”
(1966) and Bruvold and Ongerth (1969) showed that the taste sensory methodologies have been tested: sensory profile, Temporal
and the acceptance of water is linked to the total amount of min- Dominance of Sensations (TDS) and free sorting tasks were pro-
erals dissolved, which varies among bottled and tap waters. More- cessed to characterize water samples from a sensory point of view.
over, Bruvold and Gaffey (1969) showed that this total amount of First of all, classical sensory profiles have been conducted,
minerals is a good predictor of the perceived taste and quality of including descriptor generation and training of panelists. The sen-
water. sory profile is certainly the most popular methodology used to de-
Many studies have focused on off-flavors of water due to pollu- scribe products and was a natural method to discover the sensory
tion episodes. However the taste of waters that are ‘‘free of off-fla- space of drinking water.
vors” has been the subject of only few publications. Sensory Then, the Temporal Dominance of Sensations (see Labbe, Sch-
analysis of water is most often done by semi-quantitative assess- lich, Pineau, Gilbert, & Martin, 2009; Pineau et al., 2009) was ap-
ment with a methodology derived from the Quantitative Descrip- plied. TDS is a dynamic method enabling to describe the taste of
tive AnalysisÒ (Stone, Sidel, Oliver, Woolsey, & Singleton, 1974) water all along the tasting and consists in scoring the dominant
and Quantitative Flavor Profiling (Stampanoni, 1993). These meth- sensation repeatedly, until sensations end.
Finally, a free sorting task has been conducted with naïve con-
* Corresponding author at: CNRS, UMR 5171, Centre Européen des Sciences du
sumers. The sorting procedure has been successfully used to obtain
Goût, CNRS-UB-INRA, France. Tel.: +33 169935112; fax: +33 169935174. product maps (Faye et al., 2004; Giboreau, Navarro, Faye, & Dumor-
E-mail address: eric.teillet@agroparistech.fr (E. Teillet). tier, 2001; Lawless, Sheng, & Knoops, 1995; Popper & Heymann,

0950-3293/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.foodqual.2010.04.012
968 E. Teillet et al. / Food Quality and Preference 21 (2010) 967–976

1996) on various product spaces (food, beverages, textiles. . .). It Table 1


was recently considered as an alternative to Quantitative Descrip- List of the waters with their total mineralization by experiment.

tive Analysis for external preference mapping (Faye et al., 2006). Waters Total minerals Profile 1 Profile 2 TDS Sorting PSP
The free sorting task simply consists in grouping the samples (mg/L)
according to their sensory similarities. Falahee and MacRae Mont Roucous 19 X X X X
(1995) already demonstrated the efficiency of free sorting tasks ap- Epinal 59 X X
plied to the taste of water. Thann 60 X X X
Champeau 80 X
During the study, each of these methods showed drawbacks Guerlesquin 93 X
which prevent them to be used routinely. Finally, a methodology Volvic 121 X X X X
was developed in order to rapidly reveal the sensory characteris- Aquarel 152 X
tics of water, without the need for a large number of samples. This Bellegarde 167 X
Orleans 192 X X
method is named ‘‘Polarized Sensory Positioning” and is simply
Cristaline 195 X
based on the comparison of samples to water standards (the poles). Valvert 206 X X
PSP has also been developed to enable the accumulation of data Gennevilliers 225 X
across multiple studies. Vigneux 249 X
Since the water samples differ across studies which have been Dijon 290 X X X
Evian 326 X X X X
conducted successively and independently over two years, some
Evian with chlorine 326 X
caution is needed when comparing the substantive results across Thonon 329 X X X X
methods. But with this caveat in mind, comparisons and conclu- Auxerre 331 X
sions regarding the pros and cons of each methodology can still Yutz 376 X X X
Flins 460 X
be drawn. Methodologies have been compared on the basis of var-
Soissons 504 X X
ious criteria: ease of use, discrimination, interpretability, ability to Carola 668 X X
‘‘aggregate” data from several studies, ability to include verbal and Wattwiller 734 X X
hedonic assessments. Vittel 845 X X
Finally the objectives of the paper are: (1) To evaluate and com- St Amand 894 X X
Taillefine 1117 X X
pare the contribution of classical methodologies in understanding
Vauban 1244 X X X
the basics of the taste of water, and (2) To propose a new sensory Contrex 2039 X X X X
methodology well suited to the determination of the taste of water. Courmayeur 2305 X
Hepar 2664 X X X

Waters are sorted according to their mineralization and cities of tap waters are
2. Material and methods
indicated in bold type. Profile 1 refers to the sensory profile of bottled waters and
profile 2 refers to the sensory profile of tap waters.
2.1. Products

As far as possible, the waters were chosen to span the full range
of the total amount of minerals found in drinking waters in France. Table 2
Bottled waters were chosen on the same criteria from the most List of the descriptors used in experiments.

popular brands available on the French market. Tap waters were Descriptors Profile 1 Profile 2 TDS Sorting
collected from various cities of France, supplied by Lyonnaise des Sour X X X X
Eaux. The same waters were not tasted with all methodologies, Bitter X X X X
as it would have been done in a strict comparison study. But the Hard water X X
evolution of the project and the fact that tap waters are not stable Musty X X
Metallic X X X X
over time explain that the product data set is not the same for each
Metallic aftertaste X X
method. Salty X X X X
Waters were all transported into plastic bottles from a unique Cool X X X X
batch. Before tasting, tap waters were passively de-chlorinated Long aftertaste X X
by storage in a fridge for one week. The list of all the waters tasted Astringent X X X X
Tickles the throat X X X
sorted by their total mineralization is given in Table 1. Dries the mouth X X X X
For the profile of tap waters (profile 2), as French legislation rec- Minerals X
ommends a maximum of 800 mg/L and as waters between 500 and Chlorine X
800 mg/L are uncommon, the most mineralized water was only Tasteless X
Pharmacy X
504 mg/L.
Sweet X
A sample ‘‘Evian with chlorine” has been added to the products ‘‘Bottled water” X
in order to test the efficiency of PSP methodology in a space includ-
ing a sample with an off-flavor (see Section 3.2). Evian was thus
served with 0.2 mg/L of free chlorine. explicitly defined in a sensory point of view with a consensus be-
tween panelists. It was not the case in the sorting task performed
2.2. Descriptors by consumers who interpret personally the descriptors.

Profile, TDS and sorting task have been processed with a list of 2.3. Panelists and protocols
descriptors. These descriptors have been generated at the begin-
ning of the profile sessions and are listed in Table 2. PSP was not 2.3.1. Sensory profile
performed with descriptive attributes. This point is discussed in The sensory profile of bottled waters (16 products, profile 1)
part IV. and de-chlorinated tap waters (13 products, profile 2) have been
Profile and TDS have been performed with trained panelist, so processed in 2 different steps. After selection of the panelists, each
that descriptors have been selected on discrimination criteria and step included vocabulary generation and training of a panel.
E. Teillet et al. / Food Quality and Preference 21 (2010) 967–976 969

For the profile of bottled waters, 16 subjects have been selected 2.3.4. Polarized sensory positioning
on the basis of their ability in identifying and ranking the intensity In developing PSP, the idea was to elaborate a comparative
of tastes and mouth feelings. Then, subjects were asked to generate methodology that would not require the use of an excessive num-
descriptors and were trained to score them on continuous scales ber of samples for each new analysis. Other criteria were that this
for 12 sessions. Nine descriptors have been defined and selected method could be used without special training of panels and that it
for the measurements. Finally, the 16 subjects scored 9 descriptors would enable the aggregation of data from several experiments.
for 16 bottled waters in triplicate. The previous methodologies had highlighted three main tastes
The procedure has then been repeated with tap waters. Fifteen of water. In addition, one interesting feature of natural mineral
trained subjects (7 of them had participated in the first profile) waters on the market is that their mineral composition and orga-
scored 11 descriptors for 13 tap waters in duplicate. noleptic qualities are legally required to remain stable over time.
As classical methodologies highlighted three main tastes of water
(see Part III), three brands were chosen among mineral waters on
2.3.2. Temporal dominance of sensations the French market. Evian, Volvic and Vittel (called A, B and C during
The temporal dimension of the taste of water was studied using tasting sessions) were thus selected as prototypes of the three
the TDS method where each judge scored over the course of time tastes demonstrated.
the intensity of the descriptor perceived as dominant. TDS has been The proposed methodology is based on comparison with the
processed on bottled waters by the same trained panel which per- prototypes. It simply involves measuring the sensory similarities
formed sensory profile 2. In TDS, all the attributes defined during of a water sample with the three prototypes named ‘‘poles”.
the profile session were presented simultaneously on the screen Several methods for determining these similarities were inves-
with a continuous linear scale. The panelists had to click on ‘‘start” tigated. The method presented in this paper enables the distance to
when putting in mouth a sample of water and during the evalua- each pole to be obtained ‘‘independently”. The panelists were sim-
tion the subject had to select the attribute considered as dominant ply asked to provide a score relative to each pole on a continuous
and to score it. When the dominant perception changed, the sub- scale ranging from ‘‘exactly the same taste” to ‘‘totally different
ject had to score the new dominant sensation. The data collected taste”.
during the tasting of each product for each subject were the time In the example given in Fig. 1, the product 403 was judged to
when an attribute was selected as dominant; the name of the given have a taste very different from pole A and closer to that of B
attribute and its intensity. Finally, 15 subjects evaluated 13 sam- and to a lower extent of C.
ples of water with the TDS methodology. Finally, 32 consumers tasted 10 waters with this PSP
methodology.
2.3.3. Free sorting task
Whereas profile and TDS have been performed with trained 2.4. Data analysis
panels in a sensory laboratory, the free sorting task has been per-
formed with 389 naïve consumers at a conference. Anyone who at- All analyses were performed using MatlabÒ software (The
tended the conference was invited to taste waters. The free sorting MathWorks Inc., 2008) and SASÒ software (SAS Institute Inc.,
task simply consists in gathering products according to their sen- 2008). A p-value of 0.05 was used for statistical testing.
sory similarities. Two products judged similar by a consumer
would be put in the same group and two products judged different 2.4.1. Sensory profile
would be separated. The number of groups and the number of sam- With the scores given by trained panels in sensory profiles,
ples per group were free as were the gathering criteria, which may analyses of variance (ANOVAs) allowed determining the significant
vary from one consumer to another. descriptors and assessing the repeatability and the agreement
In this study, subjects were presented with the entire set of within the panels.
samples. First, the assessors had to sort waters into groups with To visualize differences between samples while taking account
similar tastes. Then, they were asked to describe each group by of subject notation heterogeneity, Canonical Variables Analysis
choosing items in a fixed list of 17 descriptors. Finally, they were (CVA) was carried out based on a multiple analysis of variance
asked whether they liked or not the taste of the waters of each (Manova) model: product + subject + product  subject. CVA axes
group. They were allowed to answer that a group could contain maximize the distances between products while minimizing the
liked and disliked waters or to indicate that they did not like nor residual variability. Confidence ellipses (90%) can be drawn around
dislike the waters of a group. each product of the CVA assuming that individual assessments can

“Compare this sample with the 3 references labeled A, B and C

Same taste Totally different


taste
A 403

Fig. 1. Assessment of the similarities between a product coded 403 and the 3 poles.
970 E. Teillet et al. / Food Quality and Preference 21 (2010) 967–976

be projected as supplementary points in the map. To learn more considered as missing. Thus, a new descriptor named ‘‘global pref-
about Manova and CVA and their advantages in comparison with erence” was defined as the sum of liking for each product and cor-
Principal Component Analysis (PCA) see for example Monrozier related to the axes.
and Danzart (2001). Finally, in order to evaluate discrimination between the prod-
ucts, confidence ellipses have been calculated thanks to a bootstrap
procedure (see Lee & Young, 1995). Random samples of 389 con-
2.4.2. Temporal dominance of sensations
sumers in the 389 participants with replacement have been pro-
First, dominance proportions were used to draw TDS curves
cessed 1000. For each new sample, MDS has been performed on
along time, enabling to understand which descriptors are domi-
the total dissimilarity matrix. After optimal rotation and reflections
nant during the tasting of water. Dominance proportions are calcu-
to fit the reference map, each new MDS coordinates enabled to as-
lated by dividing the number of citations of an attribute by the
sess the variability of the positions of the water samples. 90% Con-
number of judges and the number of replications; the higher the
fidence ellipses can be drawn.
dominance rate, the better the agreement among panelists. For
each sample of water, the curves of dominance of all the descrip-
tors were displayed on the same graph. 2.4.4. Polarized sensory positioning
Two other curves were displayed in order to help us with the Since A, B and C in the PSP procedure can be seen as products or
interpretation. The first one called ‘‘chance” represents the domi- as descriptors, there are 2 different ways to process the data.
nance rate that an attribute can obtain by chance (1/number of In one hand, if poles are considered as products, then scores gi-
attributes). The second one called ‘‘significance”, based on a bino- ven on the scale can be seen as distances between products and
mial test, expresses the smallest value of the proportion being sig- can be coded from 0 for ‘‘same taste” to 10 for ‘‘totally different
nificantly higher than the chance level. When the TDS curves are taste”. The data can be averaged by sample and techniques of
above the significance levels, they can be considered as consistent MDS unfolding (Busing, Groenen, & Heiser, 2005) can be performed
at panel level. on the mean samples  poles distance matrix. These recent tech-
In addition, the scores given to descriptors through the TDS pro- niques, still rarely used in sensory field, are useful to analyze rect-
cedure can be averaged by product and can be performed through angular distance matrices and seem to be the most natural
PCA on the products x descriptors matrix of the mean intensities technique to process PSP data.
for all subjects and all times. CVA has not been employed, because In the other hand, poles can also be considered as ‘‘global”
of the high number of missing data in the individual prod- descriptors. Data are then coded 0 for ‘‘totally different taste” to
ucts  descriptors matrices. Nevertheless, PCA gives a sensory 10 for ‘‘same taste” and can be analyzed with usual techniques
map which can be easily compared with the maps obtained with such as PCA after average. Nevertheless, since this methodology
the other sensory methodologies. has been undertaken by consumers untrained in scoring on this
scale, techniques of 3-way analysis such as Statis (Lavit, 1988)
were preferred.
2.4.3. Free sorting task
These two statistical approaches are compared in the paper.
Dissimilarities between stimuli were analyzed using non-met-
Fig. 2 summarizes the different data layouts for the unfolding
ric multidimensional scaling (MDS, Young & Hamer, 1987). For
and the Statis approaches.
any pair of samples, the number of consumers having grouped
these items in the same class was chosen as a measure of similarity
(Arabie & Boorman, 1973). Then, considering that a descriptor as- 3. Results
signed to a group was assigned to all the waters of this group, each
descriptor provided a vector of occurrences that represent the fre- 3.1. Classical methodologies
quency with which each water was so described. These occur-
rences were then correlated to the vectors of product coordinates 3.1.1. Sensory profiles
on the MDS map. This enables the interpretation of the sensory dif- The Anovas showed that 6 descriptors out of 9 were significant
ferences between samples. Finally, preference data were coded -1, for the profile of bottled waters (astringent, metallic, metallic
0 and +1 when the groups were respectively qualified as ‘‘disliked”, aftertaste, bitter, cool, salty) and only 4 out of 11 for the profile
‘‘neither liked nor disliked” and ‘‘liked”. When a group contained of tap waters (musty, bitter, acid, aftertaste duration). Moreover,
‘‘some liked and some disliked” waters, the preference data were reliability tests showed that panelists agreement and repeatability

Judge j A B C A B C
Judge 2 A B P1 C 2.3 1.7 6.8 P1 1.8 5.2 6.3
Judge 1 AP1 B 0.8 C 2.1P2 4.3 4.6 4.8 2.8 P2 3.9 6.0 2.6
Mean(X)
P1 P2 2.26.0 7 5.7…0.7
0.8 …
P2 … 5.3 3.2 …
6.0 …
… … Pp 3.4 3.1 5.4 Pp 4.3 9.2 7.3
… Pp 2.1 5.6 8.6
Pp 2.1 5.2 6.5
X
A, B, C considered A, B, C considered
10 - X as descriptors as products

Statis MDS Unfolding


Fig. 2. Diagram of data processing for j subjects, p products and 3 poles (A, B, C).
E. Teillet et al. / Food Quality and Preference 21 (2010) 967–976 971

were poor for this panel on most attributes. These results may be have a more ‘‘pronounced” taste. Waters with high mineral content
due to the difficulties of training or the difficulty to define refer- (on the bottom-right) seem to have a salty taste while waters with
ences for each descriptor on such ‘‘tasteless” products. A CAP (Con- low mineral content (on the top right) seem to be more ‘‘metallic”.
trol of Assessor Performances; Schlich, 2000) table, which The 90% confidence ellipses of Vittel and Mont Roucous have been
summarizes these reliability criteria is not presented here but is drawn, showing that the discrimination between waters is very
available in Teillet (2009). poor with the profile methodology. Particularly, there is no real
CVA of the scores of the descriptors for bottled waters (Fig. 3) discrimination between the taste of highly mineralized waters
shows a sensory space partly determined by the global mineraliza- and the taste of low mineralized waters.
tion (in parentheses in mg/L on the graphs). The profile methodol- The perceptive space of de-chlorinated tap waters (Fig. 4) has
ogy seems to separate waters with moderate mineral content on the same structure, but since there is no high mineral content
the left of the map and waters with high and low mineral content water, these samples can be separated only into water with mod-
on the right. Waters on the left are perceived tasteless while others erate mineralization and waters with low mineral contents, having

0.6

Volvic (121)
cool
0.4
Evian (326) metallic aftertaste
Valvert (206) Aquarel (152)
metallic
0.2
Dimension 2 (16.5%)

Thonon (329) St Amand (894) Mt Roucous (19)


dries the mouth
Vittel (845)
tickles the throat astringent
0 Cristaline (195) bitter
Wattwiller (450)
Contrex (2039)
-0.2 Carola (668) acid
Taillefine (1117)
salty

-0.4 Hépar (2664)


Vauban (1244)
Courmayeur (2305)

-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1


Dimension 1 (45.5%)

Fig. 3. CVA map of the sensory profile of bottled waters. The total amount of minerals is indicated between parentheses. The confidence ellipses (90%) of Vittel and Mont
Roucous are presented.

0.6

musty
Dijon (290)
0.4 salty

Bellegarde (167) Guerlesquin (93)


astringent
Vigneux (249)
0.2
Dimension 2 (14.3%)

tickles the throat


Flins (460) Thann (60) acid bitter

dries the mouth


0 hard water
Gennevilliers (225)
Auxerre (331) metallic
Epinal (59)
-0.2 Yutz (376)
Orléans (192) aftertaste duration

Champeau (80)
-0.4
cool

Soissons (504)

-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1


Dimension 1 (40.2%)

Fig. 4. CVA map of the sensory profile of tap waters. The total amount of minerals is indicated between parentheses. The confidence ellipses (90%) of Vigneux and Thann are
presented.
972 E. Teillet et al. / Food Quality and Preference 21 (2010) 967–976

a more pronounced taste. The confidence ellipses are still huge (e.g. dominated by the sensation of coolness, while others were found
Vigneux and Thann). to be less cool, e.g. Volvic, Contrex.
This first approach thus confirmed the link between mineral Except coolness, a few taste descriptors can reach the signifi-
content and taste of water. However, ANOVAs showed that the cance line, showing that the agreement within panelists is low.
panels had a weak discrimination for most descriptors, and repeat- Furthermore, the temporal dimension is apparently not required
ability was very low. In addition, the sensory profile is a time-con- for the sensory analysis of water tastes. There is apparently no real
suming and costly methodology to implement and our panelists sequentiality in the taste of water.
reported that it was very difficult to score the intensity of descrip- PCA of the scores given through the TDS procedure (Fig. 6) gives
tors in samples that were so similar from a sensory standpoint. a map which is equivalent in terms of interpretation to the one of
the sensory profile. The map is also partly structured by the total
mineral content.
3.1.2. Temporal dominance of sensations
The analysis of dominance proportions showed two main pro- 3.1.3. Free sorting task
files of water: waters with and without ‘‘cool” dominance MDS analysis showed a 4-axes space of perception but only the
(Fig. 5). Waters with moderate mineral contents, e.g. Evian, are first plan will be discussed in this paper. Fig. 7 shows the results

Fig. 5. TDS curves of Evian, Volvic and Contrex.


E. Teillet et al. / Food Quality and Preference 21 (2010) 967–976 973

obtained with the free sorting task carried out by consumers. The above all characterized by a higher mineral content perceived
sensory map is highly driven by the mineralization of the samples. and a more pronounced salty taste, while those of Mont Roucous
Tap waters have not been systematically separated from bottled and Thann are characterized by more bitterness. Finally, these four
waters as they are not isolated on the map and as they are located water samples were qualified as ‘‘acid”, ‘‘metallic” and ‘‘astringent”
according to their mineralization. Thus, bottled and de-chlorinated with a ‘‘persistent aftertaste”. These samples with a ‘‘pronounced”
tap waters seem to be part of a same space of perception. Yutz has taste are those with either the lowest or highest mineral content
a particular position in the map due to the presence of undesirable (consistent with profile and TDS results). A medium total mineral
compounds. content of about 300 mg/L thus results in a neutral taste, associ-
Then, correlations of the descriptors provide a sensory interpre- ated with the sensation of coolness of the water and finally the
tation of the map. For example, ‘‘tasteless” and ‘‘cool” descriptors overall appreciation of consumers.
were used most often by consumers to describe waters located at The confidence ellipses derived from the bootstrap procedure
the bottom left of the map (Vigneux, Dijon, Evian, Thonon. . .). (e.g. Thann) shows a better discrimination of the products than
Waters at the top or right of the map, on the other hand were in the profile and TDS procedures. It suggests that large panels of
judged to be less neutral and less cool and more pronounced in consumers can lead to a better discrimination of products than
terms of the other sensory characteristics. Hépar and Contrex are small trained panels.

0.4

Thonon (329) Evian


cool
(326)
Volvic (121)
0.2 Vittel (845)
St Amand (894) Mt Roucous (19)
Dimension 2 (13.6%)

Wattwiller (450)
0
Taillefine (1117) astringent
acid
bitter
Hépar (2664)
-0.2 Valvert (206)
metallic aftertaste
Vauban (1244)
metallic
Carola (668)
-0.4 Contrex (2039)

tickle the throat


dries the mouth
-0.6 salty

-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2


Dimension 1 (56.5%)

Fig. 6. PCA map of the scores given in TDS procedure.

2
Mont Roucous (19)
Epinal (59)
1.5
Thann (65)
bitter
1 musty pharmacy
metallic
remaining aftertaste
0.5 acid
Dimension 2 (30.6%)

Volvic (121)
astringent
salty Contrex (2039)
Orléans (192) minerals
0
Hépar (2664)

-0.5 coolness Yutz (324)


global liking Vigneux (249) Mineralization
tasteless "bottled water"
-1 Dijon (290)
Thonon (326)
-1.5
Evian (329)

-2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 3
Dimension 1 (39.8%)

Fig. 7. MDS map of the free sorting data. Tap waters are in bold type and mineralization is indicated between parentheses. The confidence ellipses (90%) of Thann and Dijon
are presented.
974 E. Teillet et al. / Food Quality and Preference 21 (2010) 967–976

3.1.4. Conclusions on classical methodologies perception. Finally, comparative methodologies such as free sort-
Lastly, all the sensory methods used showed a perceptive space ing task do not enable new samples in new analyses to be taken
which structure depends on the overall mineral composition of the into account. Indeed, as the results obtained with these methodol-
waters (Figs. 3–7), revealing three principal ‘‘water tastes”: a bitter ogies are conditioned to the set of product used, the aggregation of
and metallic taste for low mineral content waters, a neutral taste data from several studies is not possible unless all samples previ-
and a sensation of freshness for medium mineral content waters ously tested are included.
and finally the more salty taste for highest mineral content waters. PSP was thus developed as a mix between comparative and
These methods also showed that after eliminating residual chlo- monadic methodologies in order to improve sensory analysis of
rine, tap water samples were perceived in the same way as bottled water.
waters.
Each of the methodologies has its own limitations which pre- 3.2. Polarized sensory positioning (PSP)
vents it of being used routinely. The sensory profile is time-con-
suming and costly to implement and did not provide us with an The Statis compromise of the PSP data is shown on Fig. 8. In this
effective discrimination of the waters. The encountered problems approach, poles are considered as global descriptors. This mapping
may be due to the difficulty to define references for the panel train- is a correct reflection of the observed TDS gradient. In addition, one
ing or the use of continuous scales on such similar products water sample with an ‘‘off-flavor” (artificially chlorinated Evian
through a non comparative procedure. In addition, the temporal water) was detected and escaped from the mineral content
dimension provided by TDS does not seem to be present in water parabola.

Evian with chlorine (326)


0.1 Contrex (2039)

0.05
Vittel (845)
Dimension 2

Mont Roucous (19)


0
Thann (60)

Volvic (121)
Thonon (329)

-0.05
Dijon (290)
Yutz (376)

Evian (326)
-0.1
-0.1 -0.05 0 0.05 0.1
Dimension 1

Fig. 8. Statis compromise of PSP data.

-0.2

-0.15 Vittel Pole


Contrex (2039)
-0.1 Evian with chlorine (326)

Volvic Pole Vittel (845)


-0.05 Mont Roucous (19)
Volvic (121)
Dimension 2

0 Thann (60)

0.05
Yutz (376)
Thonon (329)
0.1 Dijon (290)
Evian (326)
0.15
Evian Pole

0.2
-0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25
Dimension 1

Fig. 9. Statis compromise with inclusion of the poles after dilatation (factor 2.32).
E. Teillet et al. / Food Quality and Preference 21 (2010) 967–976 975

Volvic (121)
Volvic
1.5
Mont Roucous
Mont Roucous (19) Contrex
Contrex(2039)

1
Vittel
Vittel pole
pole
Volvic pole
Volvic pole
0.5 Thann (60)
Thann
Dimension 2
Vittel
Vittel (845)
0

Evian withchlorine
Evian withchlorine(326)
with chlorine (326)
-0.5
Thonon (329) Yutz (376)
Yutz
Thonon
Dijon (290)
Dijon
-1
Evian (326)
Evian
Evian pole
Evian pole
-1.5
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
Dimension 1

Fig. 10. Unfolding map of PSP data (axes 1 and 2).

3
Evian
Evian withchlorine
withchlorine(326)
with chlorine (326)
2.5

1.5

Yutz
1 Evian(326) Yutz (367) Contrex
Evian
Dimension 2

Contrex (2039)
0.5
Mont
Mont Roucous
Roucous (19)
Volvic (121)
Volvic Vittel(845)
Vittel
Thann (60)
Thann Evian pole
Evian pole
0
Volvic
Volvic pole
pole Vittel pole
Vittel pole
-0.5
Dijon
Dijon (290)
-1
Thonon
Thonon (329)

-1.5
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
Dimension 1

Fig. 11. Unfolding map of PSP data (axes 1 and 3).

In this study, Evian, Volvic and Vittel waters were tasted as again (Fig. 10). On this plan, the off-flavor ‘‘Evian with chlorine”
poles but also as blind samples. Each pole can be supplementary seems to be closer to Evian than other poles, but this product is iso-
projected in the analysis with a distance of 0 from itself and a dis- lated on dimension 3 (Fig. 11). Thus, the first plan could represent
tance of 10 to the other poles. But Volvic, Evian and Vittel do not the ‘‘real” taste of water whereas dimension 3 points out the off-
precisely coincide with each of their corresponding pole as panel- flavors.
ists did not use the entire scoring scale. An expansion factor (deter- Even though the results obtained were consistent, validation
mined by a Procrustean analysis) was thus applied (Fig. 9), of this PSP method is currently based only on the logic of min-
showing that samples Volvic, Evian and Vittel are finally located eral content observed in prior studies. Other performance crite-
close to the corresponding poles. Thus, Statis seems to be efficient ria should be defined. The results in terms of interpretation are
in processing PSP data. nevertheless equivalent to those obtained with the different
Considering poles as products, MDS unfolding is the most rele- classical sensory methods we tested. Particularly, PSP discrimi-
vant statistical method. The analysis determined a 3 dimension nates better the tastes of low and high mineralized water than
space. In the first plan, the gradient of mineralization can be found sensory profile.
976 E. Teillet et al. / Food Quality and Preference 21 (2010) 967–976

4. Discussion – Conclusion Acknowledgements

As explained in the introduction, the experiments have not This work has been funded by Lyonnaise des Eaux and the
been designed in order to compare the sensory methodologies. Agence Nationale de la Recherche Technique (ANRT) and has been
Such a study would have required, for example, the use of the initiated by the competitive cluster VITAGORA.
same products. Nevertheless, the results obtained are expected Thanks a lot to Philippe Courcoux (ENITIAA, Nantes, France)
to emphasize the main pros and cons of the different methodol- who introduced the authors with the unfolding techniques.
ogies used.
The different classical sensory methodologies tested in this References
work have confirmed that sensory analysis of water is a difficult
task. But results suggested a continuum of the taste of water along Arabie, P., & Boorman, S. A. (1973). Multidimensional scaling of measures of
distance between partitions. Journal of Mathematical Psychology, 10, 148–203.
the total mineralization associated with 3 major kinds of taste: al- AWWA (1993). Flavor profile analysis: Screening and training of panelists. Denver, CO:
most bitter and metallic for low mineral content waters, neutral American Water Works Association (AWWA) Manual AWWA.
and fresh for medium mineral content waters and more salty for Bruvold, W. H., & Gaffey, W. R. (1969). Rated acceptability of mineral taste in water:
II. Combinatorial effects of ions on quality and action tendency ratings. Journal
highest mineral content waters. The study also highlighted the pre- of Applied Psychology, 53(4), 317–321.
dominance of coolness during the tasting of water and its link to Bruvold, W. H., & Ongerth, H. J. (1969). Taste quality of mineralized water. Journal of
the mean preference. Nevertheless, beyond the link with overall the American Water Works Association, 61, 170.
Bruvold, W. H., & Pangborn, R. M. (1966). Rated acceptability of mineral taste in
mineral content, it would be relevant to measure the contribution water: I. Journal of Applied Psychology, 50(1), 22–32.
of each mineral to the taste of natural waters. Furthermore, con- Busing, F. M. T. A., Groenen, P. J. F., & Heiser, W. J. (2005). Avoiding degeneracy in
sumer preferences are addressed and detailed in dedicated ongo- multidimensional unfolding by penalizing on the coefficient of variation.
Psychometrika, 70, 49–76.
ing works.
Falahee, M., & MacRae, A. (1995). Consumer appraisal of drinking water:
The study showed that sensory profile is certainly unsuited in Multidimensional scaling analysis. Food Quality and Preference, 6, 327–332.
sensory analysis of water, while comparative methods (free sorting Faye, P., Brémaud, D., Durand Daubin, M., Courcoux, P., Giboreau, A., & Nicod, H.
(2004). Perceptive free sorting with naïve subjects: An alternative to descriptive
task) give better results but do not enable the rapid analysis of new
mapping and a tool for sensory segmentation of consumers. Food Quality and
samples. Preference, 15, 781–792.
The method of Polarized Sensory Positioning developed offers Faye, P., Brémaud, D., Teillet, E., Courcoux, P., Giboreau, A., & Nicod, H. (2006). An
bright perspectives for routine sensory analysis of water samples. alternative to external preference mapping based on consumer perceptive
mapping. Food Quality and Preference, 17, 604–614.
Its use could be generalized to a more exhaustive study of water Giboreau, A., Navarro, S., Faye, P., & Dumortier, J. (2001). Sensory evaluation of
in France and perhaps throughout the world. Besides, PSP data automotive fabrics: The contribution of categorization tasks and non verbal
from different taste sessions and different panels have already information to set-up a descriptive method of tactile properties. Food Quality
and Preference, 12, 311–322.
been successfully aggregated by authors. Journal Officiel des Communautés Européennes (1998). Directive 98/83/CE du
PSP methodology provides promising results and should ulti- conseil du 3 novembre 1998 relative à la qualité des eaux destinées à la
mately enable the sensory characteristics of a given water to be consommation humaine. [on line]. <http://aida.ineris.fr/textes/directives/
text0507.htm>.
determined as well as limiting the number of samples to taste. Krasner, S. W., McGuire, M. J., & Ferguson, V. B. (1985). Tastes and odors: the flavor
There still remain a substantial number of questions to improve profile method. Journal of the American Water Works Association, 77(3), 34–40.
and generalize the use of the technique. Labbe, D., Schlich, P., Pineau, N., Gilbert, F., & Martin, N. (2009). Temporal
dominance of sensations and sensory profiling: A comparative study. Food
First of all from a semantic standpoint, the ‘‘Totally different
Quality and Preference, 20(3), 216–221.
taste” anchor of the scale poses a problem since this notion is Lavit, C. (1988). Analyse conjointe de tableaux quantitatifs. Paris, Masson.
inherent to each individual. In this context, a ‘‘standardization” of Lawless, H. Y., Sheng, N., & Knoops, S. S. C. P. (1995). Multidimensional scaling of
sorting data applied to cheese perception. Food Quality and Preference, 6, 91–98.
the scores of each panelist could be considered. If this is not possi-
Lee, S. M. S., & Young, G. A. (1995). Asymptotic iterated bootstrap confidence
ble, the nature of the scale could be changed. intervals. Annals of Statistics, 23, 1301–1330.
Secondly, the choice of sensory poles totally determines the re- Monrozier, R., & Danzart, M. (2001). A quality measurement for sensory profile
sults obtained. Thus, the number of poles to use and their sensory analysis: The contribution of extended cross-validation and resampling
techniques. Food Quality and Preference, 12, 393–406.
characteristics are important factors of the analysis that should be Pineau, N., Schlich, P., Cordelle, S., Mathonnière, C., Issanchou, S., Imbert, A., et al.
objectively defined. (2009). Construction of TDS curves and comparison with time intensity. Food
In addition, this PSP method provides a new type of sensory Quality and Preference, 20(6), 450–455.
Popper, R., & Heymann, H. (1996). Analyzing differences among products and
data that can be processed in different ways (PCA, MDS, Statis, panelists by multidimensional scaling. In T. Naes & E. Risvik (Eds.), Multivariate
INDSCAL, Unfolding. . .), and the choice between Statis and analysis of data in sensory science (pp. 159–184). Amsterdam: Elsevier.
Unfolding is really difficult to make even if unfolding seems to Schlich, P. (2000). CAP: Control of assessor performances. University of Missouri,
Columbia: Fifth Sensometrics.
be slightly richer in information (3 axes space). More, the perfor- Stampanoni, C. R. (1993). The quantitative flavor profiling technique. Perfumer and
mance criterion adopted for this study is based only on the struc- Flavorist, 18(6), 19–24.
ture of the sensory space studied. Other criteria should be Stone, H., Sidel, J., Oliver, S., Woolsey, A., & Singleton, R. C. (1974). Sensory
evaluation by quantitative descriptive analysis. Food Technology, 28, 24–34.
defined, such as repeatability, discrimination power of products
Suffet, I. H., Brady, B. M., Bartels, J. H. M., Burlingame, G. A., Yohe, T., & Mallevialle, J.
(via confidence ellipses) or others. These considerations will be (1988). Development of the flavor profile method into a standard method for
the topic of next works. sensory analysis of water. Water Science and Technology, 20(8–9), 1–8.
Suffet, I. H., Khiari, D., & Bruchet, A. (1999). The drinking water taste and odor wheel
Finally, the addition of descriptive (verbatim) and hedonic data
for the millennium: Beyond geosmin and 2-methylisoborneol. Water Science
via PSP methodology should be examined. and Technology, 40(6), 1–14.
In summary, the characteristics of this method make it an inter- Teillet, E. (2009). Perception, préférence et comportement des consommateurs vis-
esting candidate for other product spaces for which the concept of à-vis d’eaux du robinet et d’eaux embouteillées. Thèse de 3ème cycle. Dijon,
Université de Bourgogne.
‘‘sensory poles” can be defined, for example in perfumes or soda Young, F. W., & Hamer, R. M. (1987). Multidimensional scaling: History, theory, and
markets. applications. Hillsdsle, NJ: Lawrence Erlbaum Assoc..

You might also like