Abstract image of a woman sitting on books and studying on a laptop

Sensometrics.pdf

A
AmitSharma396539

1) Sensometrics has achieved increasing sophistication in multivariate analysis techniques and relating sensory data to other data sets. However, some argue the field focuses too much on new methods rather than assessing existing techniques. 2) Relating consumer data to sensory data through techniques like internal and external preference mapping has been widely successful. New developments include probabilistic unfolding models and accounting for attribute importance. 3) Assessing sensory panel performance is important, and techniques like repeatability plots, eggshell plots, and multivariate analyses have provided useful ways to evaluate reproducibility and discrimination. 4) Relating sensory data to shelf life and using survival analysis is an developing area that can determine consumer rejection limits over time. Time-intensity techniques also continue

Engineering
1 of 76
Download to read offline
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
Sensometrics: What have we achieved? Where are we going? Hal MacFie
Overview • Achievements – Other peoples views – Sensometrics Society – Panel performance – Relating consumer to sensory – Shelf life and TI – Metrics – Role of sensory in consumer decision making and enjoyment
What is Sensometrics? PSYCHOLOGY SENSORY MARKETING INDUSTRIAL R&D PHYSICAL and BIOLOGICAL SCIENCES FOOD SCIENCE FABRIC PERSONAL PRODUCTS STATISTICS
Current position (Popper) • Need to get existing methods integrated into practice. • lack of software, and lack of best practices. • tendency for the field to offer up new methods where several already exist or to offer twists on existing methods • little incentive to try to assess the merits of the techniques relative to one another and what their real contribution is to the bottom line for the industrial application -- namely, how much do my recommendations change depending on the approach I use.
What have we achieved? • Comments from colleagues - Wakeling – Increasing sophistication of users particularly in relation to the use of Multivariate analysis, – Surprising that sensory people can conduct and interpret a PCA without any difficulties but struggle with a multiple regression. – Unnecessary sophistication – our data is not that complex in relation to other fields
What have we achieved? • Comments from colleagues Ann Noble – communication between statisticians and sensory folks – accessibility of programs for non-stat person to use validly both resulting in use of increasingly complex multivariate programs for not only analyzing sensory data but relating it to a host of other data sets (chemical analyses, design variables etc etc)
What have we achieved? • Comments from colleagues Qannari • Regarding the past : – I'd stress the reciprocal stimulation and the cross-fertilisation of sensory analysis and multivariate analysis in general. – For instance, GPA contributed a lot in the development of free choice profiling and, on the other hand, sensory analysis has contributed a lot in making GPA very popular.
Conferences • 7 Conferences now held on a biennial basis • 100-200 attendees per conference • Papers that pass the refereeing process published in Food Quality and Preference
Classification of papers in Sensometric Special issues Topic No of Papers Sensory Panel Performance 15 Relating consumer to sensory 13 Difference tests 7 Preference Testing 6 PCA/PLS 6 Design 5 Relating sensory to instrumental 4 Multivariate Testing 3 Free choice 3 Segmentation 3 Neural Nets 3 Quality 2 Choice/ Attitudes 2 Dispersion Tests 1 Multidimensional Scaling 1 Time-Intensity 1
Assessing Sensory Panel Performance
Sensory Panel performance • Repeatability – The ability to score the same product consistently for a given attribute (Rossi 2001 quoting NB Standards) • Reproducibility – The ability to score products the same, on average, as the other panel members (Rossi 2001) • Discrimination – The ability to show a significant signal to noise ratio
Repeatability • Rossi (2001) – Showed how effective simple graphical tools can be – Here we see immediately see that there are some problem assessors with regard to repeatability
Reproducibility • Tragon (1960’s) - method to partition the assessor by sample interaction • Naes- eggshell plots (1998)
Reproducibility • Dijksterhuis (1995) – Applied PCA to each assessor on each attribute separately and compared loadings on the assumption that first PCA was signal and the others were noise
Reproducibility • Couronne (1997) – Removed assessor and sample effects and did a PCA on the matrix with assessors as variables – Here we see two groups of assessors who are almost zero correlated
Reproducibility • Qannari, MacFie and Coucoux (1998) – Used isotropic scaling and Procrustes to define performance indices that characterised how similar assessors were performing – Here Assessors F and G were found to be performing differently to the rest of the panel.
Discrimination Comparison of 1 way F ratios for assessment of wines by nose and by mouth ( Aubrey, Schlich, Issanchou and Etievant 1999)
Reproducibility and Discrimination Using two way MANOVA and CVA to compare panel performance by nose and by mouth. Thus, panelists 1, 10 and 12 seem responsible for the significant difference in the two modes observed for pepper and humus descriptors; panelist 3, for wood descriptor; panelist 11, for grilled/burned descriptor; panelists 2 and 5, for caramel and kirsch/cherry stone descriptors. ( Aubrey, Schlich, Issanchou and Etievant 1999)
Sensometrics.pdf
Assessment of Panel Performance Literature • A useful and comprehensive literature has led to a number of interesting techniques that are possibly unique and may have wider significance. • Moving forward – how can this information be used to improve panel performance – feedback?
Relating Consumer to Sensory
Consumer to Sensory • Internal Preference mapping • External Preference Mapping • Clustering/Hybrid techniques • Decision making
Fig. 4. Internal preference maps—(a) degree of liking of texture/mouthfeel and (b) degree of liking of flavor of Ranch salad dressing (n=144). Squares represent the consumers. Open circles represent the samples. L = low and H = high. F = fat and G = garlic. Internal Preference Mapping Internal preference mapping of hedonic ratings for Ranch salad dressings varying in fat and garlic flavor1 Yackinous*, Wee and Guinard FQAP 1999
Internal Preference mapping • Based on consumer data • Easy to correlate in sensory data post-hoc • Restricted to Vector if using PCA • Widely used • Significance testing , confidence intervals around but not in common use
Probabilistic unfolding models for sensory data MacKay FQAP 2001 The value of probabilistic unfolding A”typical” deterministic soln Probabilistic soln These models need to pursued.
External preference mapping
The use of PCA on sensory to derive the external set and then regression of each individual consumer on to the set using a hierarchy of models - vector, circular and elliptical has been widely and successfully applied a-adults c – children All elliptical models Descriptive analysis and external preference mapping of powdered chocolate milk Hough and Sánchez FQAP 1998
Solutions and ANOVA significance tests for the best models: Column Model F1 F2 df Model df Error SS MS R² F-ratio p-value Cl1(117) Vector -0.011 -0.065 2 7 0.490 0.245 0.076 0.287 0.759 Cl2(143) Vector 0.288 0.140 2 7 21.575 10.788 0.669 7.084 0.021 Cl3(78) Circular ideal point 2.958 -4.442 3 6 33.917 11.306 0.963 51.813 0.000 Anti-ideal Cl4(266) Vector -0.146 -0.150 2 7 7.402 3.701 0.418 2.516 0.150 In bold, p-values equal or lower to the significance level External Preference mapping Preference Map Cl1(117) Cl2(143) Cl4(266) Cl3(78) (-) Sun Gold Royal Gala Braeburn Top Red Fuji Pink Lady Granny Smith Golden Delicious Johnson's Red Gibson's Green -6 -4 -2 0 2 4 6 8 -10 -5 0 5 10 -- axis F1 --> The position of each product and the best fit model of each group are then plotted. Easy to get an estimate of a desired product’s profile using regression
Jaeger and MacFie (FQAP 2000) showed that rescaling weights according to importance for preference gave an external configuration that was closer to that obtained by internal preference mapping. Think about or go straight to the consumer! Problem: Sensory panel weighting of attributes not the same as consumers
Partial Least Squares obs e rva tions va ria ble s K N X observations variables M N Y X= sensory data and Y= liking data X= Principal components of sensory and Y= liking data X= instrumental and Y= sensory data
Lines bisecting diagram are defining segments numbered in Roman Tenenhaus, Pages, Ambroisine and Guinot (FQAP in press) Axis 1 1.0 .5 0.0 -.5 -1.0 Axis 2 1.0 .5 0.0 -.5 -1.0 PAMPRYL refr. TROPICANA refr. JOKER r.t. FRUIVITA refr. TROPICANA r.t. PAMPRYL r.t. acidity smell intensity vitamin C titer fructose glucose 96 95 94 93 92 91 90 89 88 87 86 85 84 83 82 81 80 79 78 77 76 75 74 73 72 71 70 69 68 67 66 65 64 63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 R²=1 R²=.20 I II III IV sweetening power pulp odor typicity sweetness pH after centrifugation pH before processing bitterness citric acid taste intensity saccharose
Partial Least Squares Approaches • Attractive because there is no need to do a data summarisation phase on the sensory data. • However the method is not invariant to the scaling of X and Y variables • Cross validation a useful test of predictive value • Works well when vector model predominates
Hybrid methods
Current position (Popper) • Need more error-theory based way of selecting the number of factors. • Clustering is another area that has been lacking in an error- theory. Really, how do I know if these clusters exist and what their stability is? Latent class based models try to address this, and there may be others, so again, perhaps the issue is more focus on an issue rather than new tools. • Absolutely right!
Segmentation of consumers taking account of external data. A clustering of variables approach E. Vigneau, and E. M. Qannari FQAP 2002 Has shown how to cluster respondents and project them on on to internal and external data. Very clever. Available in Senstools.
Landscape Segmentation Analysis
Glucose Fructose Saccharose Sweetening power pH before processing pH after centrifugation Titre Citric acid VitaminC Smell intensity Odour typicity Pulp Taste intensity Acidity Bitterness Sweetness ξ1 ξ2 ξ3 Judge2 Judge3 Judge96 Tenenhaus, Pages, Ambroisine and Guinot (FQAP in press) Multi-block PLS modelling
Glucose ξ1 ξ2 ξ3 Judge 2, Judge 3, Judge 96 Fructose Saccharose Sweetening power pH before process. pH after centrifugation Titre Vitamin C Citric acid Smell intensity Odour typicity Pulp Taste intensity Acidity Bitterness Sweetness -.90 .93 .08 .95 .94 -.97 - .98 .41 -.20 .71 -.64 - .93 - .95 .97 .233 (t = .36) .780 (t = 1.62) >0 >0 >0 R2 = .956 .810 (t = 2.45) Judge 2, Judge 3, Judge 96 pH before process. pH after Vitamin C Pulp Taste intensity Acidity Sweetness - - .780 (t = 1.62) R2 .810 (t = 2.45) .90 .98 Fig. 8. Estimate of the hierarchical multi-block PLS model (w and c loadings). Non significant loadings are in italic
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 Physico -Chemical Sensory CoeffCS [1]( hedonic ) Fig. 7. SIMCA-P software output, validation of the PLS regression of the hedonic score on the physico-chemical and sensory scores New PLS methods are being developed outside sensometrics that are finding use – L-PLS
Decision making aids
Fig. 4. Map showing the percentages of the judges in group I classifying a product (t1, t2) above average with the products and their characteristics X Tenenhaus, Pages, Ambroisine and Guinot (FQAP in press) -6 -4 -2 0 2 4 6 T1 -3 -2 -1 0 1 2 3 T2 PAMPRYL r.t. JOKER r.t. PAMPRYL refr. TROPICANA r.t. FRUIVITA refr . TROPICANA refr. fructose glucose saccharose sweetening power pH before processing pH after centrifugation citric acid smell intensity odour typicity pulp taste intensity acidity bitterness sweetness titre 100% 80% 60% 40% 20% 0%
Percent Consumers 8.0 10.8 13.6 16.4 19.2 22.0 24.8 27.6 30.4 33.2 36.0 38.8 41.6 44.4 47.2 50.0 52.8 55.6 58.4 61.2 -0.74 -0.37 0.00 0.37 0.74 -0.74 -0.37 0.00 0.37 0.74 A E C B D G F Percent Predicted Responses Exceeding Overall Liking of 6.5 Elliptical Ideal Point Model for Salmon PCA Dimension 2 PCA Dimension 1 This contour plotting idea, derived by Danzart, is now widely used and is available in FIZZ
Relating sensory to consumer • Crucial to the relevance of sensory in an NPD commercial environment • A large number of useful techniques • More research on significance testing, decision making aids, segmentation strategies, probabilistic methods required • Structural equation modelling may have promise
Shelf-life and Time Intensity
0 0.2 0.4 0.6 0.8 1 F(t) 0 10 20 30 40 50 Storage time Probability of rejection Shelf life: 22 h ¾The yogurt doesn’t last 22h. ¾With 22 h storage there is a 50% probability that a consumer will reject it.
References • Hough et al. 2003. Survival analysis applied to sensory shelf-life of foods. J Food Science 68: 359. • Hough et al. 2004. Determination of consumer acceptance limits to sensory defects using survival analysis. Food Quality and Preference, in press. • Calle et al. 2004. Bayesian survival analysis modeling applied to sensory shelf life of foods. 7th Sensometric Meeting, poster presentation.
Time -Intensity • Standard techniques – anova etc • Overbosch, Liu and MacFie rescaling ideas • Van beuren – PCA, Dijksterhuis PCA non centred PCA • Wendin, Janestad – modelling, parametric • Eilers and Dijksterhuis 2004 • Future – back to basics?
Fig. 6. Some examples of T–I curves according to the extended the model. The response to the individual systems are shown as thin lines. The thick line is the combined response. A parametric model for time-intensity curves Eilers, P.H.C. / Dijksterhuis, G.B., Food Quality and Preference, Apr 2004
Metrics = Measurement
Metrics • Thurstonian models • Biases • New metrics
Using the variance as well as the mean of response distributions has enabled and some clear thinking has enabled O’Mahony and colleagues to develop a valuable programme of testing that is Comparing the performance of different testing procedures The effects of warm up samples, memory , retasting etc have been tested. This programme is producing as series of useful results that resonate with sensory practitioners Worth pursuing! More workers needed!
Binomial versus Beta Binomial Ennis and colleagues have developed comprehensive software that takes the test and replicate variability into account. These tests offer potential improvements in power and sensitivity and, again, resonate with sensory professionals. My prediction is that this approach will continue to increase in popularity and should be addressed by sensometricians more widely
Biases and problems
Balancing designs for positional and carry over effects Designs widely used. Need for further research? Kunert criticisms?
0 1 2 3 4 5 6 7 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Day 1 Day 2 Day 3 Order effects for Overall Liking (Evaluation) On day 1 we used a dummy sample. The rise in scores at the end of the trial is also a common effect ct. Warm up effects What is the effect of these biases on clustering or preference mapping?
milk% 100 75 75 50 25 25 0 dark% 0 25 25 50 75 75 100 sugar gms 9 18 0 9 0 18 9 Hedonic rating without attributes with attributes Does the addition of attribute questions alter the hedonic ratings?
Is there life after the 9 point hedonic scale? • Non- English speakers generally not happy with it. • Converting feelings to numbers not natural – (Cognition vs emotion controversy)
Sensometrics.pdf
Measuring Intensity Panelist Consumer Training More Less S Q R
Triadic Elicitation Which pair of products in this triad are more similar to each other? In what way are they more similar to each other than the third?
Multi-lingual Consumer Vocabulary for Dessert Apples England* Denmark Belgium/Flemish Belgium/French Spain Acidic 20 Sur 19 Zuur 20 Acide 22 Acido 23 Sweet 15 Sod 17 Zoet 13 Sucree 14 Dulce 10 Crisp 11 Sprod 9 Krisp 3 Croquante 4 Crjiente 4 Juicy 7 Saftig 5 Suppig 5 Juteuse 10 Juguso 4 Hard/firm 7 Hard 14 Hard 11 Ferme 13 Dura 8 Soft 6 Blod 9 Zacht 5 Molle 7 Blanda 5 Bitter 7 Amere 3 Amargo 4 Cooking apple 4 Stoofappel 4 Pour la Cuisine 4 Tasteless 5 Fade 13 Insipido 11 Mealy/Floury A Melet 9 Melig 12 Farineuse 8 Harinoso 6 Coarse 5 Grov 5 Granuleuse 4 Aspera 3 Dry 4 Tor 3 Droog 3 Spongy 4 Loose Stru 5 Crumbly Korrelig 4 * Based on RGM interviews with 25 consumers in each country. Descriptors correlate significantly (r2 >0.05) with GPA consensus A Only descriptors used by 3 or more (>12%) consumers are shown
Repertory Grid is a method that elicits consumers’ response by gathering their knowledge, feelings and judgments on a particular subject for which they have experience. • Rep Grid consists of 4 primary components: Repertory Grid Method Purpose—what you want to achieve P E Q C P E Q C P E Q C Element Class/Elements—samples representative of the subject content area Constructs—bi-polar scales generated by the respondent; once generated, the respondents rate each element on each of the constructs Qualifiers—statements used to direct the respondent’s thinking • And is executed with the following key steps: 1 Provide Provide Experience Experience • Ensure relevant use experience 1 Provide Provide Experience Experience • Ensure relevant use experience 2 Create Grid Create Grid • Generate constructs • Ladder up & down 2 Create Grid Create Grid • Generate constructs • Ladder up & down 3 Rate Elements Rate Elements • Rate elements on constructs • Eliminate redundancies • Refine elements 3 Rate Elements Rate Elements • Rate elements on constructs • Eliminate redundancies • Refine elements 4 Analyze & Analyze & Interpret Interpret • PCA biplots • Other techniques-- Resistance-to- Change 4 Analyze & Analyze & Interpret Interpret • PCA biplots • Other techniques-- Resistance-to- Change
Generating and handling descriptors • A rather specialist subject of sensory • Repertory grid does it but other methods exist • The application of a text clustering statistical analysis to aid the interpretation of focus group interviews Eric Dransfield FQAP 2004 • More papers appearing – new methods needed and topic where we can make a contribution • ( maybe free choice profiling is not dead – Mostaffa)
Metrics • An important area of Sensometrics • Application of Thurstonian principles producing important results • More questioning of current practice needed • New metrics needed to replace 9 point scale • Methods to handle consumer generated descriptors needed • More researchers needed
Future (Popper) • Then there is the atheoretical stance of sensometrics. Where is the perceptual/cognitive theory that be should be striving to build? • I wish there was more theoretical work and focus on estimating parameters of a perceptual/cognitive model rather than parameters is some all-purpose statistical model. • This is what makes the Thurstonian framework and what Danny is doing attractive to me.
Fundamental modelling of sensory processes related to flavour structure activities • Not much published in FQAP or Journal of SS • Chemical Senses? • Ennis has published in this area • Expect this area to grow
Don’t forget the package
Sensometrics.pdf
Table 3. Results of the assessment of the fragrance/packaging combinations (averages from 0 to 100 and significance of the difference between the averages; n.s.=no significance; P> 0.1; all interactions were not significant) The impact of olfactory product expectations on the olfactory product experience Scharf and Volkmer FQAP 2000
A case study in relating sensory descriptive data to product concept fit and consumer vocabulary Carr, B.T. / Craig-Petsinger, D. / Hadlich, S., Food Quality and Preference, Jul 2001 Fig. 3. PLS map of fit-to-concept ratings and sensory flavor data illustrates the primary and secondary flavor drivers of fit-to-concept. The map shows that the East and West segments have different flavor drivers.
Information + Expectation Prior Expectations Product •label •package •ads •price Expectations Choice High Negative Positive Repeated use Rejection Satisfaction Low Rejection Product Use: (Sensory Properties) Confirmation/ dis-confirmation of expectations Expectations raised Expectations lowered
Success= Choice + Satisfaction + repurchase • Methods to establish the role of product performance (assessed by the user) in the total marketing mix, cheaply, easily and convincingly are required. • Choice models, expectation models, satisfaction models, repurchase models
Future (Punter OP&P) • Sensory-on-Demand. • users do not need any software except a browser and internet connection. • The logical next step is the analysis and reporting on demand, so that the the client only has to provide place, products and respondents. • Ie a 100% internet application.
Future (Van Dongen) • --Better define what it is and what it is for - I still don't know what the organization does other than the meetings every other year.... • --Help statistical folks communicate results better - write standards for showing different types of data like ASTM standards • --Educate sensory folks in statistical tools - should there be minimum statistical knowledge required for sensory folks?
Future (Mostaffa) • how to determine completely new concepts (new products with new tastes, new appearances... never heard of). • This poses a real challenge in terms of statistical designs, optimisation, statistical treatment of data because we need to go beyond the limits of the experimental universe.
Summary • Impressive literature and practical methods • Have helped sensory move towards consumer insights • Sensory measures and methods – more fundamental approaches • Make Metrics an important part of our conferences and research area • Methods for helping decision making • Keep focus on product performance • Role of sensory in satisfaction/repurchase

More Related Content

PPTX
Human-centered AI: how can we support end-users to interact with AI?
Katrien Verbert
 
PDF
Reproducibilidad mapping
Suhey Pérez
 
PPTX
Workshop on Designing Human-Centric MIR Systems
epsilon_tud
 
PPTX
Sustainable intensification indicator framework for Africa RISING
africa-rising
 
PDF
Solution Manual for Introductory Statistics 9th by Mann
pxsalehje7118
 
PPT
Company Introduction July 09
SensorySlides21
 
PPTX
Non-targeted analysis research activities at the US EPA through 2020
sethrnewton
 
PPTX
CUROP - Academic Poster Design Workshop 2014
cardiffedu
 
Human-centered AI: how can we support end-users to interact with AI?
Katrien Verbert
 
Reproducibilidad mapping
Suhey Pérez
 
Workshop on Designing Human-Centric MIR Systems
epsilon_tud
 
Sustainable intensification indicator framework for Africa RISING
africa-rising
 
Solution Manual for Introductory Statistics 9th by Mann
pxsalehje7118
 
Company Introduction July 09
SensorySlides21
 
Non-targeted analysis research activities at the US EPA through 2020
sethrnewton
 
CUROP - Academic Poster Design Workshop 2014
cardiffedu
 

Similar to Sensometrics.pdf (20)

DOCX
Business Research assignment
Abhishek Varshney
 
PDF
GIA Singapore - Introduction (Mulgan)
Global Innovation Academy
 
PPT
Application of soft computing in food processing sector
Ramabhau Patil
 
PPT
1WR RapiTests for Sensory
Alexandre Khan
 
PDF
Controlled Experiments - Shengdong Zhao
Michael Shilman
 
PPT
marketing research ch 3.ppt
amitshaha3
 
PDF
Global Principles and Practices for Hotspots Analysis and Sustainability Info...
UNEP/SETAC Life Cycle Initiative
 
PPTX
Power and sample size calculations for survival analysis webinar Slides
nQuery
 
PPTX
Best compound characterization protocol
Wenlan Hu
 
PDF
Information Scent, Searching and Stopping
DavidMaxwell77
 
PDF
1st oper as userboard workshop report
Josep Lascurain - S·G·M s.l.
 
PPT
Access versus dedicated panel: ESOMAR panel conference Dublin 2008
Kristof De Wulf
 
PDF
Human-centered AI: how can we support lay users to understand AI?
Katrien Verbert
 
PDF
Visualizing User Feedback
Eva Kaniasty
 
DOCX
1) A & B NEEDS THE APA FORMATTING AND C NEED TO BE FINISH. a. ..docx
monicafrancis71118
 
PDF
Performance Evaluation for Classifiers tutorial
Bilkent University
 
PDF
use of solvents for Gp removal.pdf
TomJMDienya
 
PDF
A/B Testing: Common Pitfalls and How to Avoid Them
Igor Karpov
 
PPTX
Analisis y diseño de experimentos avanzado_DSSMP1_introduction.pptx
ignaciozamora16
 
PDF
Ppr r101-session-3-smart
SharmiBhat
 
Business Research assignment
Abhishek Varshney
 
GIA Singapore - Introduction (Mulgan)
Global Innovation Academy
 
Application of soft computing in food processing sector
Ramabhau Patil
 
1WR RapiTests for Sensory
Alexandre Khan
 
Controlled Experiments - Shengdong Zhao
Michael Shilman
 
marketing research ch 3.ppt
amitshaha3
 
Global Principles and Practices for Hotspots Analysis and Sustainability Info...
UNEP/SETAC Life Cycle Initiative
 
Power and sample size calculations for survival analysis webinar Slides
nQuery
 
Best compound characterization protocol
Wenlan Hu
 
Information Scent, Searching and Stopping
DavidMaxwell77
 
1st oper as userboard workshop report
Josep Lascurain - S·G·M s.l.
 
Access versus dedicated panel: ESOMAR panel conference Dublin 2008
Kristof De Wulf
 
Human-centered AI: how can we support lay users to understand AI?
Katrien Verbert
 
Visualizing User Feedback
Eva Kaniasty
 
1) A & B NEEDS THE APA FORMATTING AND C NEED TO BE FINISH. a. ..docx
monicafrancis71118
 
Performance Evaluation for Classifiers tutorial
Bilkent University
 
use of solvents for Gp removal.pdf
TomJMDienya
 
A/B Testing: Common Pitfalls and How to Avoid Them
Igor Karpov
 
Analisis y diseño de experimentos avanzado_DSSMP1_introduction.pptx
ignaciozamora16
 
Ppr r101-session-3-smart
SharmiBhat
 
Ad

Recently uploaded (20)

PPTX
Software Engineering and software moduleing
deepikame6
 
PDF
ChapteR012372321DFGDSFGDFGDFSGDFGDFGDFGSDFGDFGFD
NotaProGamer
 
PPTX
Current and future trends in Computer Vision.pptx
Subramanyam Natarajan
 
PDF
Level 2 – IBM Data and AI Fundamentals (1)_v1.1.PDF
KSRIHARISASANKA
 
PDF
August -2025_Top10 Read_Articles_ijait.pdf
ijait
 
PDF
Improvement effect of pyrolyzed agro-food biochar on the properties of.pdf
LasFernandaJuchem
 
PDF
737-MAX_SRG.pdf student reference guides
ssusere2119a1
 
PDF
SMART SIGNAL TIMING FOR URBAN INTERSECTIONS USING REAL-TIME VEHICLE DETECTI...
Niraj Aarya
 
PDF
Categorization of Factors Affecting Classification Algorithms Selection
IJDKP
 
PPTX
CyberSecurity Mobile and Wireless Devices
RobinRohit2
 
PPTX
Sorting and Hashing in Data Structures with Algorithms, Techniques, Implement...
usham61
 
PPTX
Fundamentals of safety and accident prevention -final (1).pptx
Palani kumar
 
PPTX
introduction to high performance computing
JuanDavid747466
 
PPTX
communication and presentation skills 01
CdtAfrazAttaullah
 
PDF
UNIT no 1 INTRODUCTION TO DBMS NOTES.pdf
pawarbhaktiit
 
PDF
PREDICTION OF DIABETES FROM ELECTRONIC HEALTH RECORDS
ijaia
 
PDF
Exploratory_Data_Analysis_Fundamentals.pdf
Ashutosh Satapathy
 
PPTX
CURRICULAM DESIGN engineering FOR CSE 2025.pptx
Amuda3
 
PPTX
"Array and Linked List in Data Structures with Types, Operations, Implementat...
usham61
 
PPTX
Graph Data Structures with Types, Traversals, Connectivity, and Real-Life App...
usham61
 
Software Engineering and software moduleing
deepikame6
 
ChapteR012372321DFGDSFGDFGDFSGDFGDFGDFGSDFGDFGFD
NotaProGamer
 
Current and future trends in Computer Vision.pptx
Subramanyam Natarajan
 
Level 2 – IBM Data and AI Fundamentals (1)_v1.1.PDF
KSRIHARISASANKA
 
August -2025_Top10 Read_Articles_ijait.pdf
ijait
 
Improvement effect of pyrolyzed agro-food biochar on the properties of.pdf
LasFernandaJuchem
 
737-MAX_SRG.pdf student reference guides
ssusere2119a1
 
SMART SIGNAL TIMING FOR URBAN INTERSECTIONS USING REAL-TIME VEHICLE DETECTI...
Niraj Aarya
 
Categorization of Factors Affecting Classification Algorithms Selection
IJDKP
 
CyberSecurity Mobile and Wireless Devices
RobinRohit2
 
Sorting and Hashing in Data Structures with Algorithms, Techniques, Implement...
usham61
 
Fundamentals of safety and accident prevention -final (1).pptx
Palani kumar
 
introduction to high performance computing
JuanDavid747466
 
communication and presentation skills 01
CdtAfrazAttaullah
 
UNIT no 1 INTRODUCTION TO DBMS NOTES.pdf
pawarbhaktiit
 
PREDICTION OF DIABETES FROM ELECTRONIC HEALTH RECORDS
ijaia
 
Exploratory_Data_Analysis_Fundamentals.pdf
Ashutosh Satapathy
 
CURRICULAM DESIGN engineering FOR CSE 2025.pptx
Amuda3
 
"Array and Linked List in Data Structures with Types, Operations, Implementat...
usham61
 
Graph Data Structures with Types, Traversals, Connectivity, and Real-Life App...
usham61
 
Ad

Sensometrics.pdf

  • 1. Sensometrics: What have we achieved? Where are we going? Hal MacFie
  • 2. Overview • Achievements – Other peoples views – Sensometrics Society – Panel performance – Relating consumer to sensory – Shelf life and TI – Metrics – Role of sensory in consumer decision making and enjoyment
  • 3. What is Sensometrics? PSYCHOLOGY SENSORY MARKETING INDUSTRIAL R&D PHYSICAL and BIOLOGICAL SCIENCES FOOD SCIENCE FABRIC PERSONAL PRODUCTS STATISTICS
  • 4. Current position (Popper) • Need to get existing methods integrated into practice. • lack of software, and lack of best practices. • tendency for the field to offer up new methods where several already exist or to offer twists on existing methods • little incentive to try to assess the merits of the techniques relative to one another and what their real contribution is to the bottom line for the industrial application -- namely, how much do my recommendations change depending on the approach I use.
  • 5. What have we achieved? • Comments from colleagues - Wakeling – Increasing sophistication of users particularly in relation to the use of Multivariate analysis, – Surprising that sensory people can conduct and interpret a PCA without any difficulties but struggle with a multiple regression. – Unnecessary sophistication – our data is not that complex in relation to other fields
  • 6. What have we achieved? • Comments from colleagues Ann Noble – communication between statisticians and sensory folks – accessibility of programs for non-stat person to use validly both resulting in use of increasingly complex multivariate programs for not only analyzing sensory data but relating it to a host of other data sets (chemical analyses, design variables etc etc)
  • 7. What have we achieved? • Comments from colleagues Qannari • Regarding the past : – I'd stress the reciprocal stimulation and the cross-fertilisation of sensory analysis and multivariate analysis in general. – For instance, GPA contributed a lot in the development of free choice profiling and, on the other hand, sensory analysis has contributed a lot in making GPA very popular.
  • 8. Conferences • 7 Conferences now held on a biennial basis • 100-200 attendees per conference • Papers that pass the refereeing process published in Food Quality and Preference
  • 9. Classification of papers in Sensometric Special issues Topic No of Papers Sensory Panel Performance 15 Relating consumer to sensory 13 Difference tests 7 Preference Testing 6 PCA/PLS 6 Design 5 Relating sensory to instrumental 4 Multivariate Testing 3 Free choice 3 Segmentation 3 Neural Nets 3 Quality 2 Choice/ Attitudes 2 Dispersion Tests 1 Multidimensional Scaling 1 Time-Intensity 1
  • 11. Sensory Panel performance • Repeatability – The ability to score the same product consistently for a given attribute (Rossi 2001 quoting NB Standards) • Reproducibility – The ability to score products the same, on average, as the other panel members (Rossi 2001) • Discrimination – The ability to show a significant signal to noise ratio
  • 12. Repeatability • Rossi (2001) – Showed how effective simple graphical tools can be – Here we see immediately see that there are some problem assessors with regard to repeatability
  • 13. Reproducibility • Tragon (1960’s) - method to partition the assessor by sample interaction • Naes- eggshell plots (1998)
  • 14. Reproducibility • Dijksterhuis (1995) – Applied PCA to each assessor on each attribute separately and compared loadings on the assumption that first PCA was signal and the others were noise
  • 15. Reproducibility • Couronne (1997) – Removed assessor and sample effects and did a PCA on the matrix with assessors as variables – Here we see two groups of assessors who are almost zero correlated
  • 16. Reproducibility • Qannari, MacFie and Coucoux (1998) – Used isotropic scaling and Procrustes to define performance indices that characterised how similar assessors were performing – Here Assessors F and G were found to be performing differently to the rest of the panel.
  • 17. Discrimination Comparison of 1 way F ratios for assessment of wines by nose and by mouth ( Aubrey, Schlich, Issanchou and Etievant 1999)
  • 18. Reproducibility and Discrimination Using two way MANOVA and CVA to compare panel performance by nose and by mouth. Thus, panelists 1, 10 and 12 seem responsible for the significant difference in the two modes observed for pepper and humus descriptors; panelist 3, for wood descriptor; panelist 11, for grilled/burned descriptor; panelists 2 and 5, for caramel and kirsch/cherry stone descriptors. ( Aubrey, Schlich, Issanchou and Etievant 1999)
  • 20. Assessment of Panel Performance Literature • A useful and comprehensive literature has led to a number of interesting techniques that are possibly unique and may have wider significance. • Moving forward – how can this information be used to improve panel performance – feedback?
  • 22. Consumer to Sensory • Internal Preference mapping • External Preference Mapping • Clustering/Hybrid techniques • Decision making
  • 23. Fig. 4. Internal preference maps—(a) degree of liking of texture/mouthfeel and (b) degree of liking of flavor of Ranch salad dressing (n=144). Squares represent the consumers. Open circles represent the samples. L = low and H = high. F = fat and G = garlic. Internal Preference Mapping Internal preference mapping of hedonic ratings for Ranch salad dressings varying in fat and garlic flavor1 Yackinous*, Wee and Guinard FQAP 1999
  • 24. Internal Preference mapping • Based on consumer data • Easy to correlate in sensory data post-hoc • Restricted to Vector if using PCA • Widely used • Significance testing , confidence intervals around but not in common use
  • 25. Probabilistic unfolding models for sensory data MacKay FQAP 2001 The value of probabilistic unfolding A”typical” deterministic soln Probabilistic soln These models need to pursued.
  • 27. The use of PCA on sensory to derive the external set and then regression of each individual consumer on to the set using a hierarchy of models - vector, circular and elliptical has been widely and successfully applied a-adults c – children All elliptical models Descriptive analysis and external preference mapping of powdered chocolate milk Hough and Sánchez FQAP 1998
  • 28. Solutions and ANOVA significance tests for the best models: Column Model F1 F2 df Model df Error SS MS R² F-ratio p-value Cl1(117) Vector -0.011 -0.065 2 7 0.490 0.245 0.076 0.287 0.759 Cl2(143) Vector 0.288 0.140 2 7 21.575 10.788 0.669 7.084 0.021 Cl3(78) Circular ideal point 2.958 -4.442 3 6 33.917 11.306 0.963 51.813 0.000 Anti-ideal Cl4(266) Vector -0.146 -0.150 2 7 7.402 3.701 0.418 2.516 0.150 In bold, p-values equal or lower to the significance level External Preference mapping Preference Map Cl1(117) Cl2(143) Cl4(266) Cl3(78) (-) Sun Gold Royal Gala Braeburn Top Red Fuji Pink Lady Granny Smith Golden Delicious Johnson's Red Gibson's Green -6 -4 -2 0 2 4 6 8 -10 -5 0 5 10 -- axis F1 --> The position of each product and the best fit model of each group are then plotted. Easy to get an estimate of a desired product’s profile using regression
  • 29. Jaeger and MacFie (FQAP 2000) showed that rescaling weights according to importance for preference gave an external configuration that was closer to that obtained by internal preference mapping. Think about or go straight to the consumer! Problem: Sensory panel weighting of attributes not the same as consumers
  • 30. Partial Least Squares obs e rva tions va ria ble s K N X observations variables M N Y X= sensory data and Y= liking data X= Principal components of sensory and Y= liking data X= instrumental and Y= sensory data
  • 31. Lines bisecting diagram are defining segments numbered in Roman Tenenhaus, Pages, Ambroisine and Guinot (FQAP in press) Axis 1 1.0 .5 0.0 -.5 -1.0 Axis 2 1.0 .5 0.0 -.5 -1.0 PAMPRYL refr. TROPICANA refr. JOKER r.t. FRUIVITA refr. TROPICANA r.t. PAMPRYL r.t. acidity smell intensity vitamin C titer fructose glucose 96 95 94 93 92 91 90 89 88 87 86 85 84 83 82 81 80 79 78 77 76 75 74 73 72 71 70 69 68 67 66 65 64 63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 R²=1 R²=.20 I II III IV sweetening power pulp odor typicity sweetness pH after centrifugation pH before processing bitterness citric acid taste intensity saccharose
  • 32. Partial Least Squares Approaches • Attractive because there is no need to do a data summarisation phase on the sensory data. • However the method is not invariant to the scaling of X and Y variables • Cross validation a useful test of predictive value • Works well when vector model predominates
  • 34. Current position (Popper) • Need more error-theory based way of selecting the number of factors. • Clustering is another area that has been lacking in an error- theory. Really, how do I know if these clusters exist and what their stability is? Latent class based models try to address this, and there may be others, so again, perhaps the issue is more focus on an issue rather than new tools. • Absolutely right!
  • 35. Segmentation of consumers taking account of external data. A clustering of variables approach E. Vigneau, and E. M. Qannari FQAP 2002 Has shown how to cluster respondents and project them on on to internal and external data. Very clever. Available in Senstools.
  • 37. Glucose Fructose Saccharose Sweetening power pH before processing pH after centrifugation Titre Citric acid VitaminC Smell intensity Odour typicity Pulp Taste intensity Acidity Bitterness Sweetness ξ1 ξ2 ξ3 Judge2 Judge3 Judge96 Tenenhaus, Pages, Ambroisine and Guinot (FQAP in press) Multi-block PLS modelling
  • 38. Glucose ξ1 ξ2 ξ3 Judge 2, Judge 3, Judge 96 Fructose Saccharose Sweetening power pH before process. pH after centrifugation Titre Vitamin C Citric acid Smell intensity Odour typicity Pulp Taste intensity Acidity Bitterness Sweetness -.90 .93 .08 .95 .94 -.97 - .98 .41 -.20 .71 -.64 - .93 - .95 .97 .233 (t = .36) .780 (t = 1.62) >0 >0 >0 R2 = .956 .810 (t = 2.45) Judge 2, Judge 3, Judge 96 pH before process. pH after Vitamin C Pulp Taste intensity Acidity Sweetness - - .780 (t = 1.62) R2 .810 (t = 2.45) .90 .98 Fig. 8. Estimate of the hierarchical multi-block PLS model (w and c loadings). Non significant loadings are in italic
  • 39. 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 Physico -Chemical Sensory CoeffCS [1]( hedonic ) Fig. 7. SIMCA-P software output, validation of the PLS regression of the hedonic score on the physico-chemical and sensory scores New PLS methods are being developed outside sensometrics that are finding use – L-PLS
  • 41. Fig. 4. Map showing the percentages of the judges in group I classifying a product (t1, t2) above average with the products and their characteristics X Tenenhaus, Pages, Ambroisine and Guinot (FQAP in press) -6 -4 -2 0 2 4 6 T1 -3 -2 -1 0 1 2 3 T2 PAMPRYL r.t. JOKER r.t. PAMPRYL refr. TROPICANA r.t. FRUIVITA refr . TROPICANA refr. fructose glucose saccharose sweetening power pH before processing pH after centrifugation citric acid smell intensity odour typicity pulp taste intensity acidity bitterness sweetness titre 100% 80% 60% 40% 20% 0%
  • 42. Percent Consumers 8.0 10.8 13.6 16.4 19.2 22.0 24.8 27.6 30.4 33.2 36.0 38.8 41.6 44.4 47.2 50.0 52.8 55.6 58.4 61.2 -0.74 -0.37 0.00 0.37 0.74 -0.74 -0.37 0.00 0.37 0.74 A E C B D G F Percent Predicted Responses Exceeding Overall Liking of 6.5 Elliptical Ideal Point Model for Salmon PCA Dimension 2 PCA Dimension 1 This contour plotting idea, derived by Danzart, is now widely used and is available in FIZZ
  • 43. Relating sensory to consumer • Crucial to the relevance of sensory in an NPD commercial environment • A large number of useful techniques • More research on significance testing, decision making aids, segmentation strategies, probabilistic methods required • Structural equation modelling may have promise
  • 44. Shelf-life and Time Intensity
  • 45. 0 0.2 0.4 0.6 0.8 1 F(t) 0 10 20 30 40 50 Storage time Probability of rejection Shelf life: 22 h ¾The yogurt doesn’t last 22h. ¾With 22 h storage there is a 50% probability that a consumer will reject it.
  • 46. References • Hough et al. 2003. Survival analysis applied to sensory shelf-life of foods. J Food Science 68: 359. • Hough et al. 2004. Determination of consumer acceptance limits to sensory defects using survival analysis. Food Quality and Preference, in press. • Calle et al. 2004. Bayesian survival analysis modeling applied to sensory shelf life of foods. 7th Sensometric Meeting, poster presentation.
  • 47. Time -Intensity • Standard techniques – anova etc • Overbosch, Liu and MacFie rescaling ideas • Van beuren – PCA, Dijksterhuis PCA non centred PCA • Wendin, Janestad – modelling, parametric • Eilers and Dijksterhuis 2004 • Future – back to basics?
  • 48. Fig. 6. Some examples of T–I curves according to the extended the model. The response to the individual systems are shown as thin lines. The thick line is the combined response. A parametric model for time-intensity curves Eilers, P.H.C. / Dijksterhuis, G.B., Food Quality and Preference, Apr 2004
  • 50. Metrics • Thurstonian models • Biases • New metrics
  • 51. Using the variance as well as the mean of response distributions has enabled and some clear thinking has enabled O’Mahony and colleagues to develop a valuable programme of testing that is Comparing the performance of different testing procedures The effects of warm up samples, memory , retasting etc have been tested. This programme is producing as series of useful results that resonate with sensory practitioners Worth pursuing! More workers needed!
  • 52. Binomial versus Beta Binomial Ennis and colleagues have developed comprehensive software that takes the test and replicate variability into account. These tests offer potential improvements in power and sensitivity and, again, resonate with sensory professionals. My prediction is that this approach will continue to increase in popularity and should be addressed by sensometricians more widely
  • 54. Balancing designs for positional and carry over effects Designs widely used. Need for further research? Kunert criticisms?
  • 55. 0 1 2 3 4 5 6 7 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Day 1 Day 2 Day 3 Order effects for Overall Liking (Evaluation) On day 1 we used a dummy sample. The rise in scores at the end of the trial is also a common effect ct. Warm up effects What is the effect of these biases on clustering or preference mapping?
  • 56. milk% 100 75 75 50 25 25 0 dark% 0 25 25 50 75 75 100 sugar gms 9 18 0 9 0 18 9 Hedonic rating without attributes with attributes Does the addition of attribute questions alter the hedonic ratings?
  • 57. Is there life after the 9 point hedonic scale? • Non- English speakers generally not happy with it. • Converting feelings to numbers not natural – (Cognition vs emotion controversy)
  • 60. Triadic Elicitation Which pair of products in this triad are more similar to each other? In what way are they more similar to each other than the third?
  • 61. Multi-lingual Consumer Vocabulary for Dessert Apples England* Denmark Belgium/Flemish Belgium/French Spain Acidic 20 Sur 19 Zuur 20 Acide 22 Acido 23 Sweet 15 Sod 17 Zoet 13 Sucree 14 Dulce 10 Crisp 11 Sprod 9 Krisp 3 Croquante 4 Crjiente 4 Juicy 7 Saftig 5 Suppig 5 Juteuse 10 Juguso 4 Hard/firm 7 Hard 14 Hard 11 Ferme 13 Dura 8 Soft 6 Blod 9 Zacht 5 Molle 7 Blanda 5 Bitter 7 Amere 3 Amargo 4 Cooking apple 4 Stoofappel 4 Pour la Cuisine 4 Tasteless 5 Fade 13 Insipido 11 Mealy/Floury A Melet 9 Melig 12 Farineuse 8 Harinoso 6 Coarse 5 Grov 5 Granuleuse 4 Aspera 3 Dry 4 Tor 3 Droog 3 Spongy 4 Loose Stru 5 Crumbly Korrelig 4 * Based on RGM interviews with 25 consumers in each country. Descriptors correlate significantly (r2 >0.05) with GPA consensus A Only descriptors used by 3 or more (>12%) consumers are shown
  • 62. Repertory Grid is a method that elicits consumers’ response by gathering their knowledge, feelings and judgments on a particular subject for which they have experience. • Rep Grid consists of 4 primary components: Repertory Grid Method Purpose—what you want to achieve P E Q C P E Q C P E Q C Element Class/Elements—samples representative of the subject content area Constructs—bi-polar scales generated by the respondent; once generated, the respondents rate each element on each of the constructs Qualifiers—statements used to direct the respondent’s thinking • And is executed with the following key steps: 1 Provide Provide Experience Experience • Ensure relevant use experience 1 Provide Provide Experience Experience • Ensure relevant use experience 2 Create Grid Create Grid • Generate constructs • Ladder up & down 2 Create Grid Create Grid • Generate constructs • Ladder up & down 3 Rate Elements Rate Elements • Rate elements on constructs • Eliminate redundancies • Refine elements 3 Rate Elements Rate Elements • Rate elements on constructs • Eliminate redundancies • Refine elements 4 Analyze & Analyze & Interpret Interpret • PCA biplots • Other techniques-- Resistance-to- Change 4 Analyze & Analyze & Interpret Interpret • PCA biplots • Other techniques-- Resistance-to- Change
  • 63. Generating and handling descriptors • A rather specialist subject of sensory • Repertory grid does it but other methods exist • The application of a text clustering statistical analysis to aid the interpretation of focus group interviews Eric Dransfield FQAP 2004 • More papers appearing – new methods needed and topic where we can make a contribution • ( maybe free choice profiling is not dead – Mostaffa)
  • 64. Metrics • An important area of Sensometrics • Application of Thurstonian principles producing important results • More questioning of current practice needed • New metrics needed to replace 9 point scale • Methods to handle consumer generated descriptors needed • More researchers needed
  • 65. Future (Popper) • Then there is the atheoretical stance of sensometrics. Where is the perceptual/cognitive theory that be should be striving to build? • I wish there was more theoretical work and focus on estimating parameters of a perceptual/cognitive model rather than parameters is some all-purpose statistical model. • This is what makes the Thurstonian framework and what Danny is doing attractive to me.
  • 66. Fundamental modelling of sensory processes related to flavour structure activities • Not much published in FQAP or Journal of SS • Chemical Senses? • Ennis has published in this area • Expect this area to grow
  • 69. Table 3. Results of the assessment of the fragrance/packaging combinations (averages from 0 to 100 and significance of the difference between the averages; n.s.=no significance; P> 0.1; all interactions were not significant) The impact of olfactory product expectations on the olfactory product experience Scharf and Volkmer FQAP 2000
  • 70. A case study in relating sensory descriptive data to product concept fit and consumer vocabulary Carr, B.T. / Craig-Petsinger, D. / Hadlich, S., Food Quality and Preference, Jul 2001 Fig. 3. PLS map of fit-to-concept ratings and sensory flavor data illustrates the primary and secondary flavor drivers of fit-to-concept. The map shows that the East and West segments have different flavor drivers.
  • 71. Information + Expectation Prior Expectations Product •label •package •ads •price Expectations Choice High Negative Positive Repeated use Rejection Satisfaction Low Rejection Product Use: (Sensory Properties) Confirmation/ dis-confirmation of expectations Expectations raised Expectations lowered
  • 72. Success= Choice + Satisfaction + repurchase • Methods to establish the role of product performance (assessed by the user) in the total marketing mix, cheaply, easily and convincingly are required. • Choice models, expectation models, satisfaction models, repurchase models
  • 73. Future (Punter OP&P) • Sensory-on-Demand. • users do not need any software except a browser and internet connection. • The logical next step is the analysis and reporting on demand, so that the the client only has to provide place, products and respondents. • Ie a 100% internet application.
  • 74. Future (Van Dongen) • --Better define what it is and what it is for - I still don't know what the organization does other than the meetings every other year.... • --Help statistical folks communicate results better - write standards for showing different types of data like ASTM standards • --Educate sensory folks in statistical tools - should there be minimum statistical knowledge required for sensory folks?
  • 75. Future (Mostaffa) • how to determine completely new concepts (new products with new tastes, new appearances... never heard of). • This poses a real challenge in terms of statistical designs, optimisation, statistical treatment of data because we need to go beyond the limits of the experimental universe.
  • 76. Summary • Impressive literature and practical methods • Have helped sensory move towards consumer insights • Sensory measures and methods – more fundamental approaches • Make Metrics an important part of our conferences and research area • Methods for helping decision making • Keep focus on product performance • Role of sensory in satisfaction/repurchase