11th SENSOMETRICS
10th-13th July 2012, Rennes
Methodological development for sensory evaluation of
product presenting biological variability:
a case study on apple
C. Bavay, R. Symoneaux, I. Maitre, A. Kuznetsova, P.B. Brockhoff, E. Mehinagic
Introduction
•
Assessment of the sensory qualities of fruits and vegetables is of major
interest for growers, retailers and industries
•
Development of adapted sensory methodologies in the 1970s and 1980s
•
Working with fresh plant material poses special problems:
• Changes of the characteristics with time
• Non availability of suitable reference samples
• Heterogeneity/biological variability of the plant material
(Heintz and Kader, 1983; Stevens and Albright, 1980; Williams and Carter, 1977)
A Challenge
"Real variation within a given genotype may make differences among genotypes more
difficult to detect"
" One of the challenges in sensory evaluation of fruit is product variability "
Hampson et al., 2000
2
The heterogeneity of plant material in a
commercial batch
•
Difficulties still emphasised in fruits and vegetables studies
•
In studies on apples, related to variations in sensory properties within a
cultivar and even within a piece of fruit
(Allais and Létang, 2009; Azodanlou et al., 2003; Cetinkaya et al., 2006; Le Moigne et al., 2008; Lonchamp et
al., 2009; Sousa et al., 2007)
(Dever et al., 1995; Hampson et al., 2000; Harker et al., 2003; Harker et al., 2005; Seppä et al., 2012;
Symoneaux et al., 2002; Vaysse et al., 2006; Watada and Abbot, 1985)
Between
orchards
•
Between
trees
Between
apples
Between apple
pieces
However, majority of sensory studies only deal with the assessor source
of variation through the assessor effect and the product x assessor
interaction term
3
Objectives
To observe fruit-to-fruit variability occurring
within a commercial batch in sensory results
To compare 2 models of analysis of variance
(ANOVA) taking the fruit-to-fruit variability into
account or not
To recommend a methodology to get more
reliable sensory results for products presenting
variability
4
Data structure
•
•
19 assessors with 60 hours training
7 sensory attributes for texture and taste (the main drivers of preferences in
apples
•
•
•
•
•
•
•
(Daillant-Spinnler et al., 1996))
Crunchiness
Firmness
Crispness
Juiciness
Fondant
Acidity
Sweetness
•
3 cultivars : Ariane, Braeburn and Pink Lady®
•
3 random replicates
5
Apples were shared by several assessors
•
•
•
To observe the fruit-to-fruit variability : apples were cut and several
assessors (3-4) tasted the same apple
• The panel was divided into 6 groups
X 3 replicates
For each group:
 A total of 18 apples for each cultivar
6
Mean scores vary with the apples
apple 1
apple 2
apple 3
apple 4
apple 5
apple 6
apple 7
apple 8
apple 9
apple 10
apple 11
apple 12
apple 13
apple 14
apple 15
apple 16
apple 17
apple 18
7
Assessors who tasted the same apple agree
apple 1
judge 1
judge 2
judge 3
apple 14
judge 7
judge 8
judge 9
8
Accounting for fruit-to-fruit variability in
data analysis
•
Standard analysis : mixed model
(Næs, Brockhoff & Tomic, 2010)
where ε ijk ~N(0, σ² ), Assessorj ~N(0, σ² Assessor ) and Cultivar:Assessorij ~N(0, σ²
Cultivar:Assessor ); all terms are independent
•
Mixed hierarchical model including fruit effect
where εijlk ~N(0, σ² ), Assessorj ~N(0, σ²Assessor ), Cultivar:Assessorij ~N(0, σ²Cultivar:Assessor ) and
Fruitl(i) ~N(0, σ²Fruit ) ; all terms are independent
9
Accounting for fruit-to-fruit variability in
data analysis
•
Mean Squares (MS) : to observe changes in MS distribution when fruit is
added
•
Contribution to variance : to evaluate the part of variability of each
factor
•
Discrimination between cultivars (p-values of cultivar effect)
•
Analysis were done with the lmerTest package (Kuznetsova, Brockhoff &
Christensen) - R software, version 2.14.2
10
Decrease of the interaction term MS and
the residual MS
600
600
600
800
800
1000
1000
800 1000
Legend
Cultivar
Cultivar Fruit LegendFruit
Assessor Assessor
Residual
Cultivar Residual
Fruit
Cultivar:Assessor
Assessor
Residual
Cultivar:Assessor
Cultivar:Assessor
Legend
400
400
400
200
200
200
0 00
A decrease of
Residual MS
800
•
600
A decrease of
Cultivar : Assessor
interaction MS
400
•
200
Adding fruit effect
implies
0
•
1000
Mean Square
MeanCrunchiness
Square Crunchiness
Mean Square Crunchiness
Without Fruit
Without Fruit
Without Fruit
With FruitWith Fruit
With Fruit
11
Decrease of the interaction term MS and
the residual MS
Firmness
Crispness
Without Fruit
With Fruit
600
200
0
0 200
0 200
600
600
1000
1000
Crunchiness
Without Fruit
Without Fruit
Fondant
600
0
0 200
500
600
200
0
Without Fruit
With Fruit
200 400 600 800
Sweetness
With Fruit
Acidity
1000 1500
Juiciness
With Fruit
Without Fruit
With Fruit
Without Fruit
With Fruit
Comparison of mean squares
with and without fruit effect
Legend
Fruit
Redisual
0
Cultivar
Assessor
Cultivar:Assessor
Without Fruit
With Fruit
12
A large contribution of the fruit
Mean Square
Contribution
toCrunchiness
variance [%]
Crunchiness
Crunchiness
29.38 %
22.65 %
3.18 %
8.88 %
35.91 %
Legend
Cultivar
Fruit
Assessor
Residual
Cultivar:Assessor
Crunchiness
Firmness
0
28.42 %
15.2 %
4.15 %
8.01 %
44.22 %
20 40 60 80
0
•
28.42 %
15.2 %
32.73
% %
4.15
35.91 %
8.01 %
19.41 %
44.22 %
4.59 %
Juiciness
0
The fruit contribution is
large
20 040 20
60 8040
60 80
Sweetness
Juiciness
Fondant
33.42 %
6.2 %
8.96 %
35.19 20.34
%
16.23 %
•
Except for
Sweetness
32.73
%
Without Fruit
With Fruit
• Large assessor
35.91 % contribution
19.41 %
4.59 %
7.36 %
20 40 60 80
0
0
20
%
23.84 %
40 6.75
60 80%
17.29 %
31.78 %
20 40 60 80
Crispness
45.52Acidity
%
12.6 %
34.07 %
9.75 %
23.93 %
18.38 %
1.83 %
13.76 %
35.9 %
Fondant
20.34 %
23.84 %
6.75 %
17.29 %
31.78 %
7.36 %
20 40 60 80
45.52 %
12.6 %
9.75 %
18.38 %
13.76 %
Firmness
29.38 %
22.65 %
3.18 %
8.88 %
35.91 %
0
Crispness
0
20 40 60 80
4.28 %
0
20 400 60
20 408060
80
Contribution to variance [%]
Legend
Cultivar
Assessor
Cultivar:Assessor
Acidity
Fruit
Redisual
34.07 %
23.93 %
1.83 %
35.9 %
4.28 %
13
A large contribution of the fruit
Crunchiness
•
29.38 %
22.65 %
3.18 %
8.88 %
35.91 %
The fruit contribution is
•
Larger than the Assessor
contribution and Cultivar :
Assessor contribution for
Crunchiness, Firmness,
Juiciness and Fondant
•
At least larger than Assessor
contribution for Crispness
and Acidity
•
Larger than Cultivar effect
for Juiciness and Acidity
Firmness
0
20 40 60 80
28.42 %
15.2 %
4.15 %
8.01 %
44.22 %
0
Juiciness
20 40 60 80
Sweetness
 Real variation within a given
cultivar may make
differences among cultivars
more difficult to detect
33.42 %
6.2 %
8.96 %
35.19 %
16.23 %
0
45.52 %
12.6 %
9.75 %
18.38 %
13.76 %
20 40 60 80
0
Fondant
32.73 %
35.91 %
19.41 %
4.59 %
7.36 %
0
Crispness
Acidity
20.34 %
23.84 %
6.75 %
17.29 %
31.78 %
0
20 40 60 80
34.07 %
23.93 %
1.83 %
35.9 %
4.28 %
20 40 60 80
0
20 40 60 80
Contribution to variance [%]
Legend
Cultivar
Assessor
Cultivar:Assessor
Fruit
Redisual
20 40 60 80
14
The inclusion of fruit effect may change
the conclusions
Standard analysis : mixed model
Significance levels for the sensory evaluation of the three apple cultivars
Crunchiness
Firmness
Crispness
Juiciness
Fondant
Acidity
Sweetness
Cultivar
0.000 ***
0.000 ***
0.000 ***
0.001 ***
0.000 ***
0.003 ***
0.000 ***
Assessor
0.028
0.001
0.000
0.380
0.000
0.000
0.000
Cultivar:Assessor
0.999
1.000
0.304
0.053
0.996
1.000
0.185
Mixed hierarchical model including fruit effect
Significance levels for the sensory evaluation of the three apple cultivars
Crunchiness
Firmness
Crispness
Juiciness
Fondant
Acidity
Sweetness
Cultivar
0.000 ***
0.000 ***
0.000 ***
0.065 NS
0.000 ***
0.047
*
0.000 ***
Assessor
0.000
0.000
0.000
0.382
0.000
0.000
0.000
Cultivar:Assessor
0.409
0.274
0.118
0.000
0.026
0.711
0.185
Fruit
0.000
0.001
0.065
0.000
0.000
0.000
0.137
15
Conclusion
•
The fruit-to-fruit variability is an important characteristic of a batch
•
Adding the fruit effect in the analysis makes sense
•
Adding the fruit effect in the analysis can imply changes in conclusions
• E.g. erroneous conclusion about the improvement of a product
•
Recommendations
•
•
Collecting data : Each piece of fruit should be shared by several assessors
•
Analysing data : Hierarchical mixed ANOVA including fruit effect should be
applied
Perspective
•
Mixed Assessor Model
16
Thank you for
your attention
Special thanks to the panelists,
Corinne Patron and
Isabel Saillard
17