Analyzing sensory data with R
Inhoud
Part I Quantitative descriptive approaches.............................................................................................3
1 When panelists rate products according to a single list of attributes..................................................3
1.1 Data, sensory issues, and notations..............................................................................................3
1.2 In practice.....................................................................................................................................3
1.2.2 How can I asses the performance of my panel?.....................................................................4
1.2.3 How can I assess the performance of my panelists?..............................................................5
Lecture QDA and panel performance.................................................................................................6
2 When products are rated according to a single list of attributes.......................................................11
2.1 Data, sensory issues and notations.............................................................................................11
2.2 In practice...................................................................................................................................11
2.2.1 How can I get a list of the sensory attributes that structure the product space?.................11
2.2.2 How can I get a sensory profile for each product?...............................................................13
2.2.3 How can I represent the product space on a map?..............................................................13
2.2.4 How can I get homogeneous clusters of products?.............................................................14
2.3 Adding supplementary information to the product space..........................................................15
2.3.1 Introduction to supplementary information........................................................................15
3 When products are rated according to several lists of attributes......................................................15
3.1 Data, sensory issues and notations.............................................................................................15
3.2 In practice...................................................................................................................................16
3.2.1 Why can’t I analyze such a table in a classical way?.............................................................16
3.2.2 How can I get a representation of the product space based on a consensus?.....................16
Part II Qualitative descriptive approaches............................................................................................18
4 When products are depicted by comments.......................................................................................18
4.1 Data, sensory issues and notations.............................................................................................18
4.2.1 How can I approach textual data?........................................................................................19
4.2.2 How can I get an individual description of each product?...................................................20
4.3.2 How can I graphically represent the product space.............................................................20
4.2.4 How can I summarize the comments?.................................................................................21
4.3 Comparing free comments from different panels, the Rorschach test revisited.........................22
6 When products are grouped into homogeneous clusters..................................................................22
6.1 Data, sensory issues and notations.............................................................................................22
1
, 6.2 In practice...................................................................................................................................23
6.2.1 How can I approach sorting data?........................................................................................23
6.2.2 How can I get a representation of the product space?........................................................23
6.2.3 How can I fully interpret the product space?.......................................................................24
6.2.4 How can I understand the data from a panel perspective?..................................................25
7 When products are positioned onto a projective map......................................................................25
7.1 Data, sensory issues and notations.............................................................................................25
7.2 In practice...................................................................................................................................26
7.2.1 How can I approach Napping data.......................................................................................26
7.2.2 How can I represent the product space on a map?..............................................................26
7.2.3 How can I interpret the product space with the verbalization data?...................................26
7.3 The sorted Napping....................................................................................................................28
Part III Affective descriptive approaches...............................................................................................29
8 When products are solely assessed by liking.....................................................................................29
8.1 Data, sensory issues and notations.............................................................................................29
8.2 In practice...................................................................................................................................29
8.2.1 How can I approach hedonic data........................................................................................29
8.2.2 How can I identify the best product?...................................................................................32
8.2.3 How can I get homogeneous clusters of consumers?..........................................................32
8.3 Dealing with multiple hedonic variables and supplementary consumer data............................32
8.3.1 Dealing with multiple hedonic variables..............................................................................32
8.3.2 Dealing with supplementary consumer data.......................................................................33
9 When products are described by both liking and external information ‘independently’...................33
9.1 Data, sensory issues and notations.............................................................................................33
9.2 In practice...................................................................................................................................33
9.2.1 How can I explain the difference in preference using sensory data?...................................33
9.2.2 How can I evaluate the relationship between each sensory attribute and the hedonic
scores, at different levels?............................................................................................................34
9.2.3 How can I locate an optimum product within the product space?......................................36
9.3 Finding the best correspondence between the sensory and hedonic matrices, using PrefMFA. 37
10 When products are descry ed y a mix of liking and external information........................................38
10.1 Data, sensory issues and notations...........................................................................................38
10.2 In practice.................................................................................................................................39
10.2.1 How can I optimize products based on Just About Right data?..........................................40
10.2.2 How can I optimize products based on Ideal Profile Method data?...................................42
2
,Part I Quantitative descriptive approaches
1 When panelists rate products according to a single list
of attributes
1.1 Data, sensory issues, and notations
Flavor Profile Method
Spectrum
Free Choice Profiling (FCP):
Each assessor develops an individual set of sensory attributes and evaluate them according to their
personal criteria. Evaluate the attributes on a line scale (monadic sequence)
Flash Profiling:
Each assessor develops an individual set of sensory attributes and evaluate them according to their
personal criteria. Rank the samples according to intensity.
Quantitative Descriptive Analysis (QDA):
Assessors are presented with a wide range of products that present the perceptual space of interest
and are asked to individually generate a list of attributes that describe the differences among the
samples. 10-12 panelists
1. Descriptor generation (the language development) – panelists and panelleader work together
to generate a list of attributes used to describe the category of products they are studying.
2. Assessor training – assess the performance of the subjects as a whole and individually and
provide insightful and actionable information to the panel leader and to the panelists – on
which attribute can be further trained.
Discrimination – the ability to differentiate the products
Consistently (repeatable)
Consensually (in agreement)
ANOVA is used – a reference method to understand the influence of experimental factors
(the factors associated with the product effect, the panelist effect, the session effect and their
first-order intercations) on a quantitative dependent variable (sensory attributes)
3. Evaluation of samples.
All provide a sensory description of products that is as accurate as possible.
1.2 In practice
summary(donut)
Summary function – produces summaries of the results of various model fitting functions or to get an
overview of the dataset.
Set Panelist, session and rank, when they are continuous variables, as factors.
3
, donut$product<-as.factor(donut$product)
donut$panelist<-as.factor(donut$panelist)
Is the data balanced?
table(donut$product, donut$panelists)
1.2.2 How can I asses the performance of my panel?
The performance of the panel is assessed for each sensory attribute separately with an ANOVA on
each sensory attribute.
DV = product + panelists + session + product:panelist + product:session + panelist:session
res.aov<-aov(Marine~Product+Panelist+Session+Product:Panelist+Product:Session+Panelist:Session,
Perfumes.experts)
The product effect indicates whether the products are perceives as different on that
attribute. If the product effect is significant, the panel has discriminated the products with
respect to the sensory attributes of interest. The most important effect.
The panelist effect indicates whether panelists use the scale of notation similarly or not.
The session effect indicates whether the scale of notation is used consistently from one
session to the other.
The product:panelist interaction indicates whether products are perceive similarly by the
different panelists – it indicates whether there is a consensus amongst the panelists while
rating the product on the attribute of interest. If the product:panelist interaction is
significant, no consensus amongst the panelists within the panel is observed: the panelists do
not have the same perception of the products with respect to the sensory attribute of
interest. Important effect.
The product:session interaction indicates whether products are perceives similarly from one
session to the other. For a given sensory attribute of interest, it indicates whether the
attribute is used similarly from one session to the other. If the product:session interaction is
significant, the panel is not repeatable from one session to the other – measures the
repeatability. Important effect.
Use the aov function. The summary function gives the info.
The importance of a source of variability is assessed by the F-test – the ratio of two scaled sums of
squares associated with different sources of variability.
✓Product effect is significant – the panel differentiated between products.
4
Inhoud
Part I Quantitative descriptive approaches.............................................................................................3
1 When panelists rate products according to a single list of attributes..................................................3
1.1 Data, sensory issues, and notations..............................................................................................3
1.2 In practice.....................................................................................................................................3
1.2.2 How can I asses the performance of my panel?.....................................................................4
1.2.3 How can I assess the performance of my panelists?..............................................................5
Lecture QDA and panel performance.................................................................................................6
2 When products are rated according to a single list of attributes.......................................................11
2.1 Data, sensory issues and notations.............................................................................................11
2.2 In practice...................................................................................................................................11
2.2.1 How can I get a list of the sensory attributes that structure the product space?.................11
2.2.2 How can I get a sensory profile for each product?...............................................................13
2.2.3 How can I represent the product space on a map?..............................................................13
2.2.4 How can I get homogeneous clusters of products?.............................................................14
2.3 Adding supplementary information to the product space..........................................................15
2.3.1 Introduction to supplementary information........................................................................15
3 When products are rated according to several lists of attributes......................................................15
3.1 Data, sensory issues and notations.............................................................................................15
3.2 In practice...................................................................................................................................16
3.2.1 Why can’t I analyze such a table in a classical way?.............................................................16
3.2.2 How can I get a representation of the product space based on a consensus?.....................16
Part II Qualitative descriptive approaches............................................................................................18
4 When products are depicted by comments.......................................................................................18
4.1 Data, sensory issues and notations.............................................................................................18
4.2.1 How can I approach textual data?........................................................................................19
4.2.2 How can I get an individual description of each product?...................................................20
4.3.2 How can I graphically represent the product space.............................................................20
4.2.4 How can I summarize the comments?.................................................................................21
4.3 Comparing free comments from different panels, the Rorschach test revisited.........................22
6 When products are grouped into homogeneous clusters..................................................................22
6.1 Data, sensory issues and notations.............................................................................................22
1
, 6.2 In practice...................................................................................................................................23
6.2.1 How can I approach sorting data?........................................................................................23
6.2.2 How can I get a representation of the product space?........................................................23
6.2.3 How can I fully interpret the product space?.......................................................................24
6.2.4 How can I understand the data from a panel perspective?..................................................25
7 When products are positioned onto a projective map......................................................................25
7.1 Data, sensory issues and notations.............................................................................................25
7.2 In practice...................................................................................................................................26
7.2.1 How can I approach Napping data.......................................................................................26
7.2.2 How can I represent the product space on a map?..............................................................26
7.2.3 How can I interpret the product space with the verbalization data?...................................26
7.3 The sorted Napping....................................................................................................................28
Part III Affective descriptive approaches...............................................................................................29
8 When products are solely assessed by liking.....................................................................................29
8.1 Data, sensory issues and notations.............................................................................................29
8.2 In practice...................................................................................................................................29
8.2.1 How can I approach hedonic data........................................................................................29
8.2.2 How can I identify the best product?...................................................................................32
8.2.3 How can I get homogeneous clusters of consumers?..........................................................32
8.3 Dealing with multiple hedonic variables and supplementary consumer data............................32
8.3.1 Dealing with multiple hedonic variables..............................................................................32
8.3.2 Dealing with supplementary consumer data.......................................................................33
9 When products are described by both liking and external information ‘independently’...................33
9.1 Data, sensory issues and notations.............................................................................................33
9.2 In practice...................................................................................................................................33
9.2.1 How can I explain the difference in preference using sensory data?...................................33
9.2.2 How can I evaluate the relationship between each sensory attribute and the hedonic
scores, at different levels?............................................................................................................34
9.2.3 How can I locate an optimum product within the product space?......................................36
9.3 Finding the best correspondence between the sensory and hedonic matrices, using PrefMFA. 37
10 When products are descry ed y a mix of liking and external information........................................38
10.1 Data, sensory issues and notations...........................................................................................38
10.2 In practice.................................................................................................................................39
10.2.1 How can I optimize products based on Just About Right data?..........................................40
10.2.2 How can I optimize products based on Ideal Profile Method data?...................................42
2
,Part I Quantitative descriptive approaches
1 When panelists rate products according to a single list
of attributes
1.1 Data, sensory issues, and notations
Flavor Profile Method
Spectrum
Free Choice Profiling (FCP):
Each assessor develops an individual set of sensory attributes and evaluate them according to their
personal criteria. Evaluate the attributes on a line scale (monadic sequence)
Flash Profiling:
Each assessor develops an individual set of sensory attributes and evaluate them according to their
personal criteria. Rank the samples according to intensity.
Quantitative Descriptive Analysis (QDA):
Assessors are presented with a wide range of products that present the perceptual space of interest
and are asked to individually generate a list of attributes that describe the differences among the
samples. 10-12 panelists
1. Descriptor generation (the language development) – panelists and panelleader work together
to generate a list of attributes used to describe the category of products they are studying.
2. Assessor training – assess the performance of the subjects as a whole and individually and
provide insightful and actionable information to the panel leader and to the panelists – on
which attribute can be further trained.
Discrimination – the ability to differentiate the products
Consistently (repeatable)
Consensually (in agreement)
ANOVA is used – a reference method to understand the influence of experimental factors
(the factors associated with the product effect, the panelist effect, the session effect and their
first-order intercations) on a quantitative dependent variable (sensory attributes)
3. Evaluation of samples.
All provide a sensory description of products that is as accurate as possible.
1.2 In practice
summary(donut)
Summary function – produces summaries of the results of various model fitting functions or to get an
overview of the dataset.
Set Panelist, session and rank, when they are continuous variables, as factors.
3
, donut$product<-as.factor(donut$product)
donut$panelist<-as.factor(donut$panelist)
Is the data balanced?
table(donut$product, donut$panelists)
1.2.2 How can I asses the performance of my panel?
The performance of the panel is assessed for each sensory attribute separately with an ANOVA on
each sensory attribute.
DV = product + panelists + session + product:panelist + product:session + panelist:session
res.aov<-aov(Marine~Product+Panelist+Session+Product:Panelist+Product:Session+Panelist:Session,
Perfumes.experts)
The product effect indicates whether the products are perceives as different on that
attribute. If the product effect is significant, the panel has discriminated the products with
respect to the sensory attributes of interest. The most important effect.
The panelist effect indicates whether panelists use the scale of notation similarly or not.
The session effect indicates whether the scale of notation is used consistently from one
session to the other.
The product:panelist interaction indicates whether products are perceive similarly by the
different panelists – it indicates whether there is a consensus amongst the panelists while
rating the product on the attribute of interest. If the product:panelist interaction is
significant, no consensus amongst the panelists within the panel is observed: the panelists do
not have the same perception of the products with respect to the sensory attribute of
interest. Important effect.
The product:session interaction indicates whether products are perceives similarly from one
session to the other. For a given sensory attribute of interest, it indicates whether the
attribute is used similarly from one session to the other. If the product:session interaction is
significant, the panel is not repeatable from one session to the other – measures the
repeatability. Important effect.
Use the aov function. The summary function gives the info.
The importance of a source of variability is assessed by the F-test – the ratio of two scaled sums of
squares associated with different sources of variability.
✓Product effect is significant – the panel differentiated between products.
4
