You are on page 1of 21

CHAPTER 

14

Flash Profile Method


Wender L.P. Bredie, Jing Liu, Christian Dehlholm and Hildegarde Heymann

14.1 ­Introduction

Descriptive sensory analysis has been a very successful approach in determining


­relevant sensory differences between products. Methods like quantitative descrip-
tive analysis (QDA) (Stone et  al. 1974) and the more elaborate Spectrum™
Method as well as generic variations on these methods have gained popularity in
product development and research, when specific and precise descriptions of sen-
sory properties are needed (Meilgaard et  al. 2006). However, such descriptive
methods are costly and time‐­consuming and not  needed for all situations. The
flash profile method (Sieffermann 2000) is one of the attractive alternatives for
acquiring rapid sensory evaluations on products.
In conventional descriptive sensory analysis, extensive training of the panel
will in many cases prevent inconsistencies in the use of the sensory vocabulary.
Hereto, ­different approaches have been adopted. One could focus on a high
degree of training on specific sensory attributes and scales with reference materi-
als as in the Spectrum Method (Drake & Civille 2002; Meilgaard et al. 2006) or
utilize panel discussions with references for sensory attributes to reach cognitive
clarity on the sensory vocabulary between the panel members as in QDA.
Conventional descriptive approaches often follow combinations of Spectrum
Method and QDA in order to adapt to specific research goals. All of these meth-
ods clearly invest time in training of the sensory subjects in order to be able to
understand the meaning of the subsequent sensory profiles and to enable the
correlation of results to, for instance, instrumental analysis.
The flash profile method deviates from conventional profiling methods by
giving a high degree of freedom to the subjects in using their own sensory
vocabularies. Thus, the method needs less time for subjects to reach a consen-
sus vocabulary and training on the use of intensity scales. As a consequence,
the flash profile method relies strongly on subsequent data analysis to achieve

Descriptive Analysis in Sensory Evaluation, First Edition. Edited by Sarah E. Kemp,


Joanne Hort and Tracey Hollowood.
© 2018 John Wiley & Sons Ltd. Published 2018 by John Wiley & Sons Ltd.

513
514    Descriptive Analysis in Sensory Evaluation

a panel consensus space for interpretation. Whereas several studies (Dairou &
Sieffermann 2002; Delarue & Sieffermann 2004; Dehlholm et al. 2012a) have
shown that the product space is fairly robust when using the flash profile
method, interpretation of the sensory space may be more difficult. In this chap-
ter we will give an overview of the flash profile method and place it in a more
general context of methods dealing with individual vocabularies in descriptive
sensory analysis.

14.2 ­Theoretical Framework

The theoretical arguments for using individualized vocabularies as included in


the flash profile method can be founded in Kelly’s personal construct theory
(Kelly 1955). This theory was based on the idea that the construction system
used by a person to perceive, understand, predict and control the surrounding
world is composed of a finite number of dichotomous constructs. Kelly origi-
nally developed the repertory grid method (RGM) (1955) in the field of psychol-
ogy in order to establish individual constructs for objects. The main idea of this
structured interrogation technique is to get subjects to define their own con-
structs (attributes) by describing the ways in which elements and their associ-
ated meanings vary. The first application of this method in sensory science
occurred in the early 1980s. Olson (1981) introduced the idea of using this
technique to elicit the characteristics of food products. In this study, subjects
were asked to define a scale to measure the amount of each construct perceived
in the objects. Thus, each individual proposes his own constructs and scales for
evaluating the objects. Applications of the RGM have been extensive in the
fields of cognitive science (Feixas et al. 2010), food science in order to generate
sensory product attributes (Berg & Rumsey 2006) and to study consumer con-
ceptualizations in market research (Frost & Braine 1967; Swahn et  al. 2010;
Veinand et al. 2011; Yan et al. 2011).
The flash profile method promotes an individual approach in eliciting a lan-
guage to rank product perceptions but is much less rigid than the RGM in the
selection of ­product attributes. In the flash profile method, the panel consensus is
determined subsequently and it is not necessarily clear what the attributes from
the individual panel members mean in the consensus space. As can be seen in
Table 14.1, understanding of the true meaning of a sensory attribute can be diffi-
cult and confounded with different distinctions.
In the ideal case, the panel members use a term in relation to their product
evaluation in the same way and a true consensus is obtained. But in other cases,
discrepancies may occur, including correspondence, where panel members use a
different term for the same percept; conflict, where panellists use the same term
for a different percept; and contrast, where panellists use a different term for
different percepts. Gaines and Shaw (1993) highlighted the importance of
Flash Profile Method    515

Table 14.1  Different ways in which the sensory vocabulary of a panel can relate


to the distinctions between the sensory stimuli.

Sensory terminology

Same Different

Consensus Correspondence
Stimulus distinctions

People use sensory attributes for People use different sensory attributes
Same

distinctions in the same way for the same distinction


Conflict Contrast
Different

People use same sensory attributes People differ in sensory attributes and
for different distinctions distinctions

Source: adapted from Lorho (2010)

r­ ecognizing each of these situations to better measure the level of commonality


and idiosyncrasy between individual conceptual ­systems. In conventional profile
methodology, group sessions with the products ­available are very important
because they stimulate discussion and motivate panellists to explain their own
perceptions to the panel members. But even in this approach a true panel con-
sensus is not guaranteed.
Free choice profiling (FCP) (Williams & Arnold 1984; Williams & Langron 1984)
emphasized the use of individual vocabularies. This method gained popularity in
the 1980s and 1990s, and was used with analytical sensory panels but mostly to
generate product‐relevant vocabularies with consumers. FCP has lost popularity as
a descriptive analysis method mainly because of the difficulty in providing a rigid
background for interpreting the descriptive sensory space. More details on this
method are given in Chapter 13.
The flash profile method can be seen as a more recent variant of the free
choice ­profiling method. The flash profile combines the individual attribute elici-
tation of the FCP method with comparative ranking of samples on these attrib-
utes. The comparison of all products at the same time, with the possibility to go
back to previous assessments, is claimed to remove the need for product famil-
iarization and training with the attri­butes. The process of elicitation and training
is thereby drastically reduced. By making use of expert subjects, a flash profile
can be obtained over a very few sessions.
No reports have been published on how the flash profile method compares to
free choice profiling and other individual vocabulary methods such as the RGM.
However, since the flash profile relies on experienced or expert subjects, it may be
expected that these subjects have already a considerable frame of reference for
naming sensory p ­ erceptions. The flash profile method has been compared with
conventional descriptive sensory profiling on products that are typical for an
industry situation of brand comparisons (Delarue & Sieffermann 2004). The study
516    Descriptive Analysis in Sensory Evaluation

concluded that the flash profile could discriminate between products as well as
the conventional profiling method. However, the semantic structure was more
complex for the flash profile method. In another study on fruit jams (Dairou &
Sieffermann 2002), where the product differences were rather obvious, a similar
conclusion was reached. The flash profile was significantly faster than conven-
tional profiling, fairly robust and thus lowered the cost of the evaluation. Other
studies have shown similar findings (Blancher et al. 2007; Dehlholm et al. 2012a).

14.3 ­Overview of the Flash Profile Method

14.3.1 Aim
The flash profile aims to provide a rapid and reliable sensory profile of a set of
products according to their major sensory differences. The method relies on indi-
vidual vocabularies and comparative assessments to shorten the time needed to
perform the analysis.

14.3.2 Panel
The flash profile was originally developed to be carried out by experienced pan-
ellists. The panel could be sensory experts and/or professionals who have well‐
established sensory experiences with the products. For the number of subjects,
the original flash profile method used eight panel members (Dairou & Sieffermann
2002) but later ­applications of the method reported 6–12 panel members to be
sufficient (Albert et  al. 2011; Delarue & Sieffermann 2004; Dehlholm et  al.
2012a), whereas Delarue (2014) stated that a panel of 4–5 members is the mini-
mum to yield a stable configuration on sensory product differences.
Recent studies have also explored the application of the flash profile method
with consumer panels. In these studies, 24–50 consumers were considered suf-
ficient (Moussaoui & Varela 2010; Varela & Ares 2012, 2014). Depending on the
objectives of the study, the number of consumers could even range to 200
(Ballay et al. 2006). Consumer panels, however, require special explanation on
the use of the method and instruction on using product attributes. The interpre-
tation of the attribute space from consumer panels may also be difficult due to
the extensive and unclear use of sensory terminology.

14.3.3 Process
A schematic overview of the main steps in the flash profile method is given in
Figure  14.1. The sessions typically take place in computerized sensory booths
without interaction with the other panel members. A whole product set is pre-
sented simultaneously to the panel after they are given a brief outline of the test
procedure.
In the first session, each subject creates his/her own provisional list of attrib-
utes when confronted with the series of stimuli/products. The subjects are asked
to list the sensory characteristics that best describe the differences between the
Flash Profile Method    517

Selection of Data analysis and


subjects (n = 5 – 10) interpretation

Typical flash profile process

Confrontation with Adjustment of


Flash profile sample
stimuli and structured individual vocabulary
ranking in replicated
individual vocabulary with attributes from
sessions (min. 2)
development other panel members

All stimuli present All stimuli present All stimuli present


~0.5–1 hour ~0.5 hour ~1–3 hours

Figure 14.1  Schematic overview of the steps in the original flash profile method.

samples and are instructed to avoid the use of hedonic terms. To assist in vocabu-
lary development, the subjects are instructed to group terms according to sense
modality such as appearance, sound, t­ exture, taste, etc. They can re‐examine the
products as often as they wish and take as much time as they need. Subjects are
instructed to pause between exposures to the stimuli/products to minimize carry‐
over effects such as adaptation. The attribute generation process lasts about
30–45 min and involves assessment of all the samples. The time spent will vary
depending on the number and kind of products to be evaluated.
In a subsequent session, a pooled list of all generated attributes is provided
to the subjects. They are allowed to modify their own list accordingly to make
sure that all relevant dimensions are included by checking the respective
­samples. A break will be necessary for the panel leader to update the individual
vocabularies.
In the final sessions the subjects rank the samples according to their own
vocabulary one attribute at the time in order of the perceived level of intensity.
The original flash profile method suggests three replicate sessions (Dairou &
Sieffermann 2002). This r­epetition step has also been carried out in FCP studies
to enable evaluation of panel performance (Costell et  al. 1995; Deliza et  al.
2005). But the number of replicates may vary depending on the type of stimuli/
products. The number of samples to be evaluated with the flash profile method
also varies and will determine the number of replicates to be used. The method
allows for ties, that is, allocating the same rank number to samples that have the
same perceived intensity.

14.3.4  Alternative Approaches


It has been noted that the flash profile can result in complex vocabularies. In
order to shorten and better define the lists of attributes, an alternative to the
flash profile method was proposed by Lorho (2010). He combined elements from
FCP, RGM and flash profile in one method called individual vocabulary profiling
(IVP). The main improvement of the IVP method, compared to the flash profile,
518    Descriptive Analysis in Sensory Evaluation

Selection of Sensory
subjects profiling

Individual vocabulary development process

First elicitation Second elicitation Training phase


Familiarization
Word list generation Vocabulary generation Practice with attributes
to stimuli
Repertory Grid (flash profile approach) Vocabulary refinement

Diad and triad Multiple stimulus Multiple stimulus


All stimuli present
presentations comparison method comparison method
~1 hour
~1 hour ~1 hour ~0 to 2 hours

Figure 14.2  Schematic overview of the steps in an extended flash profile method for acoustic
panels proposed by Lorho (2010).

was the investment of slightly more time for developing and defining the indi-
vidual vocabularies. This circumvents the main shortcoming of the flash profile
method of the use of attributes with limited interpretative value. Nevertheless,
the IVP method still benefits from faster speed, less cost and not being limited to
trained panels, compared to traditional descriptive analysis. The IVP approach
(Lorho 2010) is schematically presented in Figure 14.2. The steps involved are
familiarization with the stimuli, two sessions on individual attribute elicitation
and a training phase for vocabulary refinement. The method has been success-
fully introduced for sound quality evaluations.
Another alternative to the flash profile is ranking descriptive analysis (RDA),
­proposed by Richter et al. (2010). This method utilized a consensus vocabulary
and ranking of samples with a relatively large trained panel. The RDA method
involves four steps:
1 selection of the most frequently observed attributes
2 definition of the selected attributes by the panel
3 development of a consensus evaluation procedure with the panel
4 ranking of the samples for all the selected attributes in one replicate session.
In this way, RDA can be considered as a compromise between the flash profile
and QDA. Despite the larger number of assessors required, RDA has the advan-
tage of lower costs due to the requirement for fewer sessions than QDA. Similar
to the flash profile, a limited number of samples can be included in RDA, because
of the ranking exercise. The idea of applying a ranking test instead of scaling in
the description of a product has been suggested by Rodrigue et al. (2000).

14.3.5  Practical Considerations


When setting up the flash profile, the experimenter should reflect on a number
of ­considerations mostly related to the time spent on the practical part of the
analysis. When a large set of samples is chosen for the analyses, the individual
Flash Profile Method    519

vocabulary development becomes a more time‐consuming task. A relevant short-


cut can be to present a subset of samples to the subjects. Omitting individual
samples is possible, for example where the sample lies within a construct spanned
by other samples. This is often the case in experimental studies, while it is more
uncertain in sets of market‐related samples.
After the first individual vocabulary development session, the individual lists
of vocabularies need to be distributed in some way to the whole panel. As it
might take some time to copy all lists, a break before the next vocabulary devel-
opment session is a good idea. After the final vocabulary development, the panel
leader must take into account the way in which he/she records the individual
assessments on the individual attributes ranks. This practical procedure is differ-
ent from an ordinary consensus p­rofiling and is dependent on the choice of
sensory software. The whole flash profile can, with some effort, be completed
within one day, of course depending on the nature of products and the aim of
the study. But if one wishes to avoid stressing subjects and also wishes to per-
form repetitions, the test is recommended to run over several days.
The descriptive result of the flash profile might include a great number of
attributes, blurring the results plot considerably relative to one from a conven-
tional descriptive profile. One must allocate sufficient time for analysis and inter-
pretation of flash profile data. The method sets higher demands for the proficiency
of the standard analyst. If lemmatization (grouping of similar attributes) is car-
ried out, one must be consistent and report the process. For example, one could
group the attributes of the panel according to broad and more refined categories,
as one does when building a sensory wheel. This will allow grouping of attributes
with a similar meaning. For instance, the term ‘fruity’ can be linked to subcate-
gories of lemon, banana, pear, etc. Since the flash profile does not generate a
panel consensus, the panel leader needs enough knowledge on the products to
be able to generate a post hoc grouping of sensory attributes by looking at the
consensus configuration plots (see section  14.4). For instance, categories like
‘fruity’, ‘banana’, ‘ester‐like’ and ‘sweet’ may be grouped to a ‘sweet‐fruity’
dimension for a particular product, but the panel leader will require sufficient
experience and product knowledge to group terms together in a meaningful way.

14.3.6  Statistical Analysis


14.3.6.1  Data Collection
As the evaluation is performed as rank ordering, transferring the ranking posi-
tions on the paper sheet into scores is the first step in data analysis. An example
of a ranking sheet is presented in Figure  14.3. The product with the lowest
intensity is given a score of 1, and then the score is increased according to the
ranking order of each product. In this way, the rank numbers are ascending
according to the increasing of the sensory intensity (rank position) for the attrib-
ute. The rank data are then easily given interpretation in multivariate analysis,
since higher rank values correspond to the higher ­perceived sensory intensities.
520    Descriptive Analysis in Sensory Evaluation

Name: Jack

Attribute 1: Sweet
1 C 2.5 4 5 B 6.5
Low D E A G F High

Attribute 2: Floral B
1 D 3 E 5.5 7
Low A G F C High

Attribute 3: Banana
1 A 2.5 D 4.5 6 7
Low B G E F C High

Figure 14.3  Example of a flash profile ranking sheet and data coding (only three attributes are
represented).

Assessor 1
Product Sweet Floral Banana …
A 5 1 2.5
B 6.5 3 1
C 2.5 7 7
D 1 3 4.5
E 2.5 5.5 4.5
F 6.5 5.5 6
G 5 3 2.5

Σ 28 28 28

Figure 14.4  Example of a data table with rank numbers from a flash profile corresponding to
the ranking sheet in Figure 14.3. The rank numbers are presented here for only one replicate.

For products with equal rank positions (tied samples), a mean rank value is
given. An example is given in Figure  14.4 showing a data table with rank
­numbers corresponding to the ranking sheet in Figure 14.3. In this way indi-
vidual matrices for each assessor (products × attributes) are built. In case rank
numbers are assigned manually into data tables, the sums of the rank values for
each attribute should be ­identical (see Figure 14.3).

14.3.6.2  Univariate Analysis


Since the set of attributes is often unique for each assessor in flash profile data,
univariate analysis can be used to look at attribute consistency for sensory asses-
sors. The flash profile ranking data could be analysed by non‐parametric tests
such as the Friedman test. For instance, the Friedman two‐way analysis of vari-
ance (ANOVA) by ranks could be used to check for consistency in the replica-
tions. Those attributes showing a poor statistical reliability could be omitted from
Flash Profile Method    521

Sweet
Product
Replicate 1 Replicate 2
A 4 5
B 6.5 6.5
C 2.5 1.5
D 1 1.5
E 2.5 3
F 6.5 6.5
G 5 4

Figure 14.5  Example of a Spearman’s test result when evaluating the panellists’ reproducibility
between two replicates. The Spearman rank correlation coefficient is: r = 1 – 6 × Sum (V2) / (N ×
(N2 – 1)). V is the differences between the ranks of both variables: V = Rank (R1) – Rank (R2);
N is the number of products. In this example, r = 0.9375.

further multivariate analysis. Alternatively, one‐way ANOVA may be used


(Lawless & Heymann 2010). One‐way ANOVA (product as the main factor) is
usually run on the flash profile data for evaluating the panellists’ performance.
Attributes that are found not to discriminate the products significantly might be
excluded from the list of assessors. Spearman’s correlation coefficients can also
be calculated and grouped for each assessor in order to evaluate the panellist’s
repeatability (Gkatzionis et  al. 2013; Price et  al. 2014; Rason et  al. 2006). An
example of Spearman’s correlation test result when evaluating reproducibility
between two replicates is shown in Figure 14.5. One could easily run this test in
the XLSTAT software. However, the use of univariate analysis such as Spearman’s
correlation test for the flash profile data also depends on the project aim and
nature of the samples, etc. If one prefers to take all assessors’ measures into
account, all attributes without any exclusion could be used for the following
multivariate analysis.

14.3.6.3  Multivariate Analysis


A data analysis method that is particularly well suited for flash profile is general-
ized Procustes analysis (GPA) (Gower 1975), similarly to the analysis of FCP
data. The GPA uses translation, rotation and isotropic scaling of the individual
product sets in order to maximize the consensus configuration. It can handle
different sizes of response ­variables on the sample objects for each subject. When
an optimal consensus configuration is found, the panel average position of the
products can be related to the sensory ­attributers used by all subjects. Also, for
attributes that individuals have in common, the agreement between individuals
in the product space can be visualized. This approach is very similar to the analy-
sis of FCP data but is based on ordinal (rank order) data instead of scale data.
Before processing of the data by GPA, the flash profile results are normally
averaged over the replicates. The structure of the data matrix will be the same as
presented in Figure  14.4, but with the values replaced by the averaged rank
522    Descriptive Analysis in Sensory Evaluation

numbers over the ­replicates. The flash profile data thus consist of blocks of vari-
ables with different sizes, one block for each assessor. However, the GPA can also
be run on the separate data sets of the replicates. The results from the GPA are
divided over different dimensions with decreasing importance. As a validation
tool, GPA uses a permutation test as diagnostic, by comparing the experimental
configuration with a random configuration. Only GPA results with configura-
tions deviating in the permutation test should be given interpretation. The first
two dimensions of the GPA are often sufficient to show the consensus configura-
tion for the samples and the correlation space for the attributes.
An alternative way of analysing flash profile data is by multiple factor analysis
(MFA) (Dehlholm et al. 2012a). MFA (Escofier & Pagès 1994) is in this case based
on PCA as the data are quantitative and allow grouping of the variables and sam-
ples. MFA on flash profile data will give a similar result as GPA. However, GPA
seems to be better suited for finding the optimal consensus space in individual
scaling data. As a consequence, it may give somewhat higher explained variance
than MFA. Both GPA and MFA analyses can be done, for instance, in Senstools
software (OP&P), XLSTAT (Addinsoft) and R (Ihaka & Gentleman 1996; R
Development Core Team 2010) using the functionality of the add‐on packages
SensoMineR and FactoMineR. The code for running GPA using FactoMineR has
been reported by Varela and Ares (2014). The code for running MFA with confi-
dence ellipses, for example showing the 95% probability ellipses around the sam-
ple scores, has been reported by Dehlholm et  al. (2012b). XLSTAT provides
tutorials on how to run GPA and MFA. The program runs as an add‐in for
Microsoft Excel.
It must be noted that the flash profile rank data are ordinal in nature, mean-
ing that the distances between rank numbers do not necessarily correspond to
the differences in perceived intensities between the samples. A sample may be
much higher in a sensory intensity than all the other samples, but receive only
a rank number that is one integer higher. For the interpretation of the results
from the data analysis, one should bear this in mind.

14.3.7  Advantages and Disadvantages


The main benefit of the flash profile is its speed while still achieving a reasonable
accuracy. The whole profiling can be completed in a few sessions, which would
take much longer in the more elaborate conventional profiling methods, like
QDA, Spectrum Method, etc. The simultaneous presentation of products is very
efficient because subjects do not need any familiarization phase with samples.
From the start, they can generate attributes that discriminate and are relevant for
the product. No training phase is needed because no agreement is required
between subjects about terms, definitions and evaluation procedures and because
the use of ordinal scales is intuitive.
Another advantage of the flash profile is the ease of experimental set‐up. It
needs significantly fewer sessions than a conventional descriptive profiling.
Flash Profile Method    523

There are no problems with matching panellist schedules as each session can be
arranged ­individually. All the samples are prepared before the sessions and the
subjects have all the samples at the same time. In practice, this means that the
experimenter does not need to be present.
A third advantage of the flash profile method in comparison to the conven-
tional ­profiling methods is the representation of diverse viewpoints by the sub-
jects. They choose their vocabulary according to their own sensitivity and
perception. The combination of each subject description reflects different points
of view depending on the importance given by the subject to each sensory
modality. This combination may enrich the sensory description of the product.
However, in some cases it may be difficult for the panel to evaluate all the
samples simultaneously due to fatigue. This is a limitation of the method but in
practice, 10 or more samples can be tested, depending on the kind of stimuli/
products. For instance, Tarea et al. (2007) reported a flash profile with six trained
assessors who successfully examined the sensory texture of 49 samples of pear
and apple purées. However, it is worth noting that the subjects declared that the
task was really tedious. In this particular case, the assessors used 4–7 attributes
and the rankings were found easy because of diversity of the product set.
Other limitations of the flash profile may arise in cases where precise tem-
perature control over the samples is needed or when evaluating the stability of
products over time, due to the need for simultaneous sample presentation.
Compared to conventional profiling where terminology is not a major concern,
the flash profile could also lead to difficulty in interpretation of sensory charac-
teristics, due to a large number of terms and lack of definitions and evaluation
procedures. Furthermore, the flash profile t­ erminology may not be easily reused
by another panel, which is easier in conventional profiling where the panel is
trained with reference standards for sensory attributes.
In cases where sensory attributes of products need to be correlated to instru-
mental measurements, it is unclear whether to recommend the use of the flash
profile. Lassoued et al. (2008) successfully correlated the instrumental texture
of bread with the flash profile using MFA to combine the data sets. Also, a
recent study showed that flash profile analysis of Stilton blue cheese with dif-
ferent levels of yeast inoculum could be used to correlate to volatile analysis
using partial least squares regression (Price et  al. 2014). However, such
approaches deserve careful consideration. At present, there is not enough evi-
dence to suggest that the flash profile can be used as an alternative for
­conventional profiling in such correlation studies. In particular, the lack of sen-
sory ­definitions and focus on the main sensory differences may lead to impre-
cise and incorrect interpretation. As shown by Dehlholm et al. (2012a) in their
study on liver paté, the flash profile provided the best discrimination among fast
sensory profiling methods (sorting, Global Napping® and Partial Napping®), but
still missed important information on sample differences, which were perceived
using conventional profiling.
524    Descriptive Analysis in Sensory Evaluation

The training of subjects on a sensory vocabulary is a time‐consuming and


costly effort, and may not be needed in all situations. Particularly in circum-
stances where one can afford to have a lower degree of semantic interpretability
at a lower cost and higher speed, alternative sensory profiling methods have
become a first choice. The flash ­profile method ranks products in relation to each
other for one attribute at a time and is more closely related to conventional pro-
filing methods than perceptual mapping methods. These latter methods, such as
Napping and ultra‐flash profiling (Pagès 2003, 2005), need more cognitive pro-
cessing of multiple sensory impressions at the same time, to make a judgement
of where to place a product in relation to other products. Such methods are even
faster but also more difficult to interpret since the response strategies may be
diverse and are much less controlled. Dehlholm et al. (2012a) clearly showed
the advantage of the flash profile in comparison to Napping and ultra‐flash pro-
file, but argued that structured approaches to Napping, such as Partial Napping,
best combined speed and accuracy in rapid sensory profiling. Even though one
could claim that such methods are based on more holistic judgements, it is too
early to conclude that perceptual mapping techniques provide better profiling
data than the flash profile method.

14.3.8 Applications
The flash profile has been applied to many different food products, including
jams (Dairou & Sieffermann 2002), dairy products (Delarue & Sieffermann
2004), commercial fruit purées (Tarea et al. 2007), jellies (Blancher et al. 2007),
bread (Lassoued et al. 2008), hot beverages (Moussaoui & Varela 2010), lemon
iced teas (Veinand et  al. 2011), fish nuggets (Albert et  al. 2011), liver paté
(Dehlholm et al. 2012a), meat (Ramírez‐Rivera et al. 2012), carrot snacks (Dueik
et al. 2013), blue cheese (Price et al. 2014), chocolates (Rezende et al. 2015) and
wine (Liu et al. 2016). Also, it has been successfully applied for non‐food prod-
ucts such as hall acoustics (Lokki et al. 2011).
Many applications of the flash profile have been reported and it has become
clearer in which situations the method is particularly well suited. In cases where
a precise sensory definition of the products is not needed, but flexibility and
rapidity of test results are demanded, the flash profile has the advantage above
conventional profiling, for instance, when working with products with a short
shelf‐life, or when fast product insights are in demand in the early stages of
product development. But when quicker results are needed, other methods like
projective mapping may be preferred.

14.4 ­Case Studies

14.4.1  Case Study 1: One‐Shot Analysis


The suitability and robustness of the flash profile were tested in a sensory evalu-
ation of smoked fresh cheese.
Flash Profile Method    525

14.4.1.1 Sample
The cheese samples were prepared from kvark (1% fat; 12% protein; Arla
Foods, Denmark) which was manually smoked in small portions in a table
smoker at 30 °C. Different wood types (beech flakes, common alder and hay),
additive (fresh orange peel or not) and smoking times were chosen to span a
space for measuring sensory differences (Table 14.2).
Furthermore, for each replicate session of the flash profile, two production
batches from the same smoking treatment (3 min) were included (common
alder with orange peel (CP3‐1 and CP3‐2)). This was done to assess the variation
due to batch production, since the reliability of the benchtop smoking process
was unknown.

14.4.1.2 Method
A trained sensory panel consisting of six subjects (three men and three women)
­participated in the test. The sensory panel followed a standard flash profile pro-
tocol with two replicated sample evaluations.
The flash profile was conducted in three sessions where all of the nine sam-
ples were presented. In the first session, the assessors were given an explanation
about the procedure. Then, all assessors were asked to list the sensory character-
istics that best described the differences among the samples and were instructed
to avoid the use of hedonic terms. The process of attributes generation lasted
around 30 min. After a short break the assessors were given a pooled attribute
list derived from all panel members on the whiteboard. They were allowed to
modify their list by adding or deleting the attributes. In the final two sessions,
assessors were required to rank the cheese samples according to their own
updated attribute list in the sensory booths. They needed to place the code of
samples on a line scale anchored at the left side representing low and right side
for high. Ties were allowed.

Table 14.2  Selected samples for the flash profile of smoked fresh cheese.

Sample code Wood type Smoke additive Smoking time Batch number
(minutes)

B1 Beech flakes None 1 1


B3 Beech flakes None 3 1
BP3 Beech flakes Orange peel 3 1
C1 Common alder None 1 1
C3 Common alder None 3 1
CP3‐1 Common alder Orange peel 3 1
CP3‐2 Common alder Orange peel 3 2
H5 Hay None 5 1
HP3 Hay Orange peel 3 1
526    Descriptive Analysis in Sensory Evaluation

14.4.1.3  Data Analysis


The score sheets from the assessors were collected and the rank numbers for all
nine samples and six assessors were entered in Excel. Rank numbers in the data
set were not averaged for the two replicates. Thus each replicate was treated as
a unique sample. In this way, the variation due to the replicates in comparison
to the variation due to the smoking treatment could be assessed.
The data were analysed by GPA in Senstools using the Procrustes free choice
module, since the flash profile data structure is similar to free choice profiling
data. The permutation test showed a total variance accounted value of 75%,
which was much higher than the 5% upper value of the total variance accounted
for of 71%. Therefore, the GPA results contained significant systematic informa-
tion. Subsequently, the scree plot, showing the consensus and residual variance,
was interpreted and two dimensions were selected, since the third dimension
only added little extra information.

14.4.1.4  Result and Discussion


In the GPA result, 33% of the variance was explained in the first dimension and
the second dimension explained 13%. Figure  14.6 shows the GPA consensus
space for the samples. Dim 1 clearly showed that the addition of orange peel
gave marked differences in flavour of the beech smoked fresh cheeses (B3 versus

C1
1
B1
Dim 2 (13%)

B3
0
H5
BP3
C3
HP3
CP3-1 CP3-2

–1

–2

–3 –2 –1 0 1 2 3
Dim 1 (33%)

Figure 14.6  GPA consensus product map of the flash profile analysis of smoked fresh cheese.
Sample codes: B, beech wood smoke; C, common alder smoke; H, hay smoke; P, orange peel
smoke additive; 1, 1 min smoking; 3, 3 min smoking; 5, 5 min smoking; codes 3‐1 and 3‐2
signify batch replicates for smoking at 3 min.
Flash Profile Method    527

BP3). Smoking with hay for 5 min gave a similar effect as smoking for 3 min for
beech and common alder, which was expected due to the lower smoke conden-
sate onto the cheese. Furthermore, the smoking time of 1 min versus 5 min for
beech and common alder smoking was clearly shown on Dim 2. The robustness
of the flash profile could be assessed from the variation between the sensory
replicates, which appeared to be limiting the discrimination between the H5, B3
and C3 samples as well as the BP3 and HP3 samples. However, the replicate vari-
ation was not similar for all samples, indicating that the panel could more easily
describe the less complex samples H1 and B1 due to the weak smoke flavour
and BP3 and HP3 due to the strong orange peel flavour. However, the batch vari-
ation (CP3‐1 versus CP3‐2) was smaller than the sensory replicate variation.
When looking into the attribute spaces, it was clearly observed that the indi-
vidual vocabularies were extensive. They varied from 19 to 36 sensory attrib-
utes. Some interpretation could be  given to the individual sensory spaces, for
example for the orange peel samples (BP3, CP3 and HP3) ‘citrus’, ‘fruity’, ‘man-
darin’, ‘apricot’, etc. terms were used whereas the most severely smoked cheeses
(H5, B3 and C3) were dominated by ‘smoked’, ‘burnt’, ‘tarmac’ attributes, which
were absent for the least smoked samples.
Overall, the flash profile analysis could show main differences between the
samples, but the sensory replicate variation was relatively large for some of the
samples. In this particular case, the sensory panel was expected to perform more
accurately if time for training on developing and using the individual sensory
vocabularies had been allowed. The case showed that one has to carefully con-
sider the use of fast descriptive analysis in relation to the merits of speed, costs
and need for accuracy.

14.4.2  Case Study 2: Small Sensory Differences


The reliability of the flash profile was investigated in a sensory evaluation of a
model wine system. In this study, flavours were added to a neutral wine, thus
controlling the expected sensory differences between the wine samples. The
samples were designed in such a way that the differences were noticeable but
relatively small. Therefore, the applicability of the flash profile for samples with
small sensory differences was investigated. The case was part of a larger study to
investigate the reliability of the flash profile and other rapid methods in com-
parison to conventional profile analysis (Liu et al. 2016).

14.4.2.1 Sample
A model wine system was designed where three different volatile compounds,
benzaldehyde, isopentyl acetate and 2‐phenylethanol, were added into a plain
white wine (Pinot Blanc, 11.5% v/v) at two different concentrations. In order to
investigate the performance of the panels, two blind replicates of benzaldehyde
high and 2‐phenylethanol low were included. The list and details including
odour descriptions are shown in Table 14.3.
528    Descriptive Analysis in Sensory Evaluation

Table 14.3  Samples used in the flash profile of a model wine system.

Flavour compounds added Level of addition (μL/L) Odour descriptors†

Benzaldehyde High* 8 Almond


Benzaldehyde Low 2 Almond
Isopentyl acetate High 5 Banana
Isopentyl acetate Low 1 Banana
2‐Phenylethanol High 100 Rose/floral/spicy
2‐Phenylethanol Low* 10 Rose/floral/spicy
Base wine – A plain white wine

*The two wines were used as blind replicates.


†Odour descriptors reported in Flavornet (www.flavornet.org/f_kovats.html).
Source: Liu et al. (2016). Reproduced with permission of Elsevier.

14.4.2.2 Method
The flash profile was carried out by a trained sensory panel composed of
eight assessors. The flash profile consisted of three sessions in which the nine
samples were presented simultaneously using a similar flash profile proce-
dure as described in section 14.4.1.2.

14.4.2.3  Data Analysis


A GPA analysis was carried out on the average data from the two replicates of
the flash profile using XLSTAT (version 2014.5.03, Addinsoft). For the analysis,
the Gower method was selected and the default permutation test with 300 per-
mutations. The program provides two options for running the GPA, namely
Gower and Commandeur. Commandeur is used when missing values occur in
the data set.

14.4.2.4  Result and Discussion


A total of 34 flavour attributes were chosen by the panel and each assessor used
a vocabulary varying from six to 13 flavour descriptors. Clearly, in this relatively
simple product, fewer attributes were used than in case study 1 on smoked
cheeses. However, given the three flavour modifications of the wine, the sensory
vocabulary was still ­relatively extensive.
The GPA consensus product map of the flash profile is displayed in
Figure 14.7. The first two dimensions accounted for 57% of the explained var-
iance (39% and 17%, dimensions 1 and 2 respectively). The wines with the
same added volatile compound were close to each other. The wines with added
volatiles in high concentration were located furthest from the centre of the
plot and stretched the plot into three different directions. The wines with low
concentrations of added volatiles were located closest to the centre and near
to  the basic wine. Benzaldehyde and isopentyl acetate groups were in the
Flash Profile Method    529

Iso Strong
4

Iso Low
2

Dim 2 (17%) Benz Low


0
Phen Low
Basis wine
Benz Strong
–2

–4 Phen Strong

–6
–6 –4 –2 0 2 4 6
Dim 1 (39%)

Figure 14.7  GPA consensus product map of the flash profile analysis of wine samples.
Benzaldehyde, isopentyl acetate and 2‐phenylethanol are abbreviated to Benz, Iso and 2‐
Phen, respectively. Benz High and 2‐Phen Low were each presented twice in each flash profile
replicate. The line connecting these sample pairs gives an indication of the reliability of the
evaluation of the weak (Phen low) versus the stronger (Benz) attribute intensity.

­ositive side of dimension 1 and dimension 2, respectively, and the


p
2‐­phenylethanol group was in the negative part of dimension 1 and 2.
However, all three samples with 2‐phenylethanol were close to the basic wine,
indicating that the  panel had difficulties distinguishing the wines with this
compound. This was ­confirmed in the conventional profile analysis of the
wines, where the difference between the wines with added 2‐phenylethanol
was just significant. The flavour ­differences of the wines with benzaldehyde
and isopentyl acetate additions were more marked (Liu et al. 2016).
The positioning of the two pairs of blind replicates is considered as a way to
check the panel performance and examine the accuracy of the methods. The
blind duplicates of benzaldehyde high were very close to each other, indicating
good reproducibility among the panel. As expected, the blind replicate for the
2‐phenylethanol low was judged less reliable. This was probably due to greater
confusion in the ranking of the samples with more similar sensory properties
(basic wine, 2‐phenylethanol low). If the wines were experimental wines where
one did not know the sensory difference beforehand, one may have concluded
that the wines were different. Hence, when the sensory differences among prod-
ucts are clearly distinguishable, one can use the flash profile method but when
the differences are relatively small, confusion in ranking may occur which could
lead to spurious conclusions. In this case the flash profile method might not be a
good choice.
530    Descriptive Analysis in Sensory Evaluation

14.5 ­Summary

The flash profile is quick to generate a sensory vocabulary, and subsequently


each ­subject uses his/her own vocabulary to rank all samples on their relative
differences for the sensory attributes. There is no need to conduct extensive
training on the product properties. However, the comparative assessment limits
the method to the number of samples that can be tested. Furthermore, since the
flash profile relies on subsequent alignment of the products × attributes space,
this method leads to a lower degree of precision, especially when the attribute
space is complex, for example when determining the flavour of foods. In cases
where the attribute space is limited, as for texture or sound, the flash profile has
the advantage of generating fast descriptive results on the main sensory differ-
ences between products. The ambiguous use of attributes in the flash profile
may have an advantage in cases where the generation of multiple attributes is
desired, such as for vocabularies for communication with consumers. In com-
parison to other rapid sensory mapping methods like Napping, results from the
flash profile contain quantitative information on the intensity of the attributes.
Thus the flash profile method is a good compromise in situations where a rapid
respond is needed but enough time is available to run two panel sessions.

14.6 ­Future Developments

One of the difficulties of the flash profile is the dependence of the results on the
capability of individual subjects to name their perceptions. Some assessors may
also drop some terms because they may undervalue their own sensory abilities.
The strict separation of panel members in order to adopt their individual vocab-
ularies may be made more flexible, allowing more interactions between the
panellists. This  would lead to more comprehensive sensory vocabularies and
still maintain the rapidness of the method. Instead of presenting individually
the list of every term at the second session in the flash profile, it could be con-
sidered to hold a group session or apply structured interviewing techniques
(Lorho 2010). In a joint session, every subject could present their own list of
terms and explain the reasons for their choices. This would give significant
improvements on the sensory concepts and may also lead to more comprehen-
sive lists of attributes without compromising the individual vocabularies.

­References
Albert, A., Varela, P., Salvador, A., Hough, G. & Fiszman, S. (2011) Overcoming the issues in the
sensory description of hot served food with a complex texture. Application of QDA®, flash
profiling and projective mapping using panels with different degrees of training. Food Quality
& Preference, 22(5), 463–473.
Flash Profile Method    531

Ballay, S., Warrenburg, S., Sieffermann, J.M., Glazman, L. & Gazano, G. (2006) A New
Fragrance Language: Intercultural Knowledge and Emotions. Paper presented at the 24th
IFSCC Congress – Integration of Cosmetic Sciences, Osaka, Japan.
Berg, J. & Rumsey, F. (2006) Identification of quality attributes of spatial audio by repertory grid
technique. Journal of the Audio Engineering Society, 54(5), 365–379.
Blancher, G., Chollet, S., Kesteloot, R., Nguyen Hoang, D., Cuvelier, G. & Sieffermann, J.M.
(2007) French and Vietnamese: how do they describe texture characteristics of the same
food? A case study with jellies. Food Quality & Preference, 18, 560–575.
Costell, E., Trujillo, C., Damasio, M.H. & Duran, L. (1995) Texture of sweet orange gels by free‐
choice profiling. Journal of Sensory Studies, 10, 163–179.
Dairou, V. & Sieffermann, J.M. (2002) A comparison of 14 jams characterized by conventional
profile and a quick original method, the flash profile. Journal of Food Science, 67(2), 826–834.
Dehlholm, C., Brockhoff, P.B., Meinert, L., Aaslyng, M.D. & Bredie, W.L.P. (2012a) Rapid descrip-
tive sensory methods – comparison and validation of free multiple sorting, partial napping,
napping, flash profiling and conventional profiling. Food Quality & Preference, 26, 267–277.
Dehlholm, C., Brockhoff, P.B. & Bredie, W.L.P. (2012b) Confidence ellipses: a variation based on
parametric bootstrapping applicable on multiple factor analysis results for rapid graphical
evaluation. Food Quality & Preference, 26, 278–280.
Delarue, J. (2014) Flash profile, its evolution and uses in sensory and consumer science.
In: Delarue, J., Lawlor, B. & Rogeaux, M. (eds) Rapid Sensory Profiling Techniques. Sawston:
Woodhead Publishing, pp. 121–151.
Delarue, J. & Sieffermann, J.M. (2004) Sensory mapping using flash profile. Comparison with
a conventional descriptive method for the evaluation of the flavour of fruit dairy products.
Food Quality & Preference, 15, 383–392.
Deliza, R., MacFie, H. & Hedderley, D. (2005) The consumer sensory perception of passion‐fruit
juice using free‐choice profiling. Journal of Sensory Studies, 20(1), 17–27.
Drake, M.A. & Civille, G.V. (2002) Flavor lexicons. Comprehensive Reviews in Food Science and Food
Safety, 2, 33–40.
Dueik, V., Marzullo, C. & Bouchon, P. (2013) Effect of vacuum inclusion on the quality and the
sensory attributes of carrot snacks. LWT – Food Science and Technology, 50(1), 361–365.
Escofier, B. & Pagès, J. (1994) Multiple factor‐analysis (Afmult package). Computational Statistics
and Data Analysis, 18, 121–140.
Feixas, G., Montebruno, C., Dada, G., del Castillo, M. & Compan, V. (2010) Self construction,
cognitive conflicts and polarization in bulimia nervosa. International Journal of Clinical and Health
Psychology, 10(3), 445–457.
Frost, W.A.K. & Braine, R.L. (1967) The application of the Repertory Grid technique to prob-
lems in market research. Commentary, 9, 161–175.
Gaines, B.R. & Shaw, M.L.G. (1993) Knowledge acquisition tools based on personal construct
psychology. Knowledge Engineering Review, 8, 49–85.
Gkatzionis, K., Hewson, L., Hollowood, T., Hort, J., Dodd, C.E.R. & Linforth, R.S.T. (2013) Effect
of Yarrowia lipolytica on blue cheese odour development: flash profile sensory evaluation of
microbiological models and cheeses. International Dairy Journal, 30, 8–13.
Gower, J.C. (1975) Generalized Procustes analysis. Psychometrika, 40(1), 33–51.
Ihaka, R. & Gentleman, R. (1996) R: a language for data analysis and graphic. Journal of
Computational and Graphical Statistics, 5, 299–314.
Kelly, G. (1955) The Psychology of Personal Constructs. New York: Norton.
Lassoued, N., Delarue, J., Launay, B. & Michon, C. (2008) Baked product texture: correlations
between instrumental and sensory characterization using Flash Profile. Journal of Cereal
Science, 48(1), 133–143.
Lawless, H.T. & Heymann, H. (2010) Sensory Evaluation of Food: Principles and Practices, 2nd edn.
New York: Springer.
532    Descriptive Analysis in Sensory Evaluation

Liu, J., Grønbeck, M.S., di Monaco, R., Giacalone, D. & Bredie, W.L.P. (2016) Performance of
Flash Profile and Napping with and without training for describing small sensory differences
in a model wine. Food Quality & Preference, 48, 41–49.
Lokki, T., Patynen, J., Kuusinen, A., Vertanen, H. & Tervo, S. (2011) Concert hall acoustics
assessment with individually elicited attributes. Journal of the Acoustical Society of America,
130(2), 835–849.
Lorho, G. (2010) Perceived quality evaluation and application to sound reproduction over
headphones. PhD thesis, Department of Signal Processing and Acoustics, Aalto University,
Espoo, Finland.
Meilgaard, M., Civille, G.V. & Carr, B.T. (2006) Sensory Evaluation Techniques, 4th edn. Boca
Raton: CRC Press.
Moussaoui, K.A. & Varela, P. (2010) Exploring consumer product profiling techniques and their
linkage to a quantitative descriptive analysis. Food Quality & Preference, 21(8), 1088–1099.
Olson, J.C. (1981) The importance of cognition processes and existing knowledge structures for
understanding food acceptance. In: Criteria of Food Acceptance. Zurich: Forester Publishing.
Pagès, J. (2003) Direct collection of sensory distances: application to the evaluation of ten white
wines of the Loire Valley. Sciences des Aliments, 23, 679–688.
Pagès, J. (2005) Collection and analysis of perceived product inter‐distances using multiple fac-
tor analysis: application to the study of 10 white wines from the Loire Valley. Food Quality &
Preference, 16, 642–649.
Price, E.J., Linforth, R.S., Dodd, C.E. et  al. (2014) Study of the influence of yeast inoculum
concentration (Yarrowia lipolytica and Kluyveromyces lactis) on blue cheese aroma develop-
ment using microbiological models. Food Chemistry, 145, 464–472.
R Development Core Team (2010) R: A Language and Environment for Statistical Computing.
3‐900051‐07‐0. Vienna: R Foundation for Statistical Computing.
Ramírez‐Rivera, E.J., Camacho‐Escobar, M.A., Garcia‐Lopez, J.C. et al. (2012) Sensory analysis
of Creole turkey meat with flash profile method. Open Journal of Animal Sciences, 02(01),
1–10.
Rason, J., Léger, L., Dufour, E. & Lebecque, A. (2006) Relations between the know‐how of
small‐scale facilities and the sensory diversity of traditional dry sausages from the Massif
Central in France. European Food Research and Technology, 222(5–6), 580–589.
Rezende, N.V., Benassi, M.T., Vissotto, F.Z., Augusto, P.P.C. & Grossmann, M.V.E. (2015) Mixture
design applied for the partial replacement of fat with fibre in sucrose‐free chocolates.
LWT – Food Science and Technology, 62, 598–604.
Richter, V.B., de Almeida, T.C.A., Prudencio, S.H. & de Toledo Benassi, M. (2010) Proposing a
ranking descriptive sensory method. Food Quality & Preference, 21(6), 611–620.
Rodrigue, N., Guillet, M., Fortin, J. & Martin, J.F. (2000) Comparing information obtained from
ranking and descriptive tests of four sweet corn products. Food Quality & Preference, 11, 47–54.
Sieffermann, J.M. (2000) Le profil flash: un outil rapide et innovant d’analyse sensorielle
descriptive. In: Proceedings of AGORAL 2000, XII rencontres ‘L’innovation: de l’idée au suc-
cess’, Montpellier, France, pp. 335–340.
Stone, H., Sidel, J.L., Oliver, S., Woolsey, A. & Singleton, R.C. (1974) Sensory evaluation
by quantitative descriptive analysis. Food Technology, 28, 24–34.
Swahn, J., Ostrom, A., Larsson, U. & Gustafsson, I.B. (2010) Sensory and semantic language
model for red apples. Journal of Sensory Studies, 25(4), 591–615.
Tarea, S., Cuvelier, G. & Sieffermann, J.M. (2007) Sensory evaluation of the texture of 49 com-
mercial apple and pear purees. Journal of Food Quality, 30, 1121–1131.
Varela, P. & Ares, G. (2012) Sensory profiling, the blurred line between sensory and consumer
science. A review of novel methods for product characterization. Food Research International,
48, 893–908.
Flash Profile Method    533

Varela, P. & Ares, G. (2014) Novel Techniques in Sensory Characterisation and Consumer Profiling.
Boca Raton: CRC Press.
Veinand, B., Godefroy, C., Adam, C. & Delarue, J. (2011) Highlight of important product char-
acteristics for consumers. Comparison of three sensory descriptive methods performed by
consumers. Food Quality & Preference, 22(5), 474–485.
Williams, A.A. & Arnold, G.M. (1984) A new approach to sensory analysis of foods and bever-
ages. In: Adda, J. (ed.) Progress in Flavour Research. Proceedings of the 4th Weurman Flavour
Research Symposium, Elsevier, Amsterdam, pp. 35–50.
Williams, A.A. & Langron, S.P. (1984) The use of free choice profiling for the examination of
commercial ports. Journal of the Science of Food and Agriculture, 35, 558–568.
Yan, W., Chen, C.H. & Chang, W. (2011) A functional‐commercial analysis strategy for product
conceptualization. Expert Systems with Applications, 38(8), 9879–9887.

You might also like