Sie sind auf Seite 1von 24

DK9848_C057.

qxd 04/Nov/05 04:46 PM Page 57-1

57 Sensory Science: Methodology


Joel L. Sidel and Herbert Stone
Tragon Corporation, Redwood City

CONTENTS

I. Introduction and Overview ..................................................................................................................................57-3


A. Historical Background ................................................................................................................................57-3
B. Defining Sensory Evaluation ......................................................................................................................57-3
C. Classification of Sensory Methods ..............................................................................................................57-4
D. Sensory vs. Instrumental Testing ................................................................................................................57-4
II. Sensory Evaluation Facilities ..............................................................................................................................57-5
A. Testing ..........................................................................................................................................................57-5
1. Reception Area ......................................................................................................................................57-5
2. Test Kitchen Area ..................................................................................................................................57-5
3. Sensory Booths ....................................................................................................................................57-6
B. Panel Training ..............................................................................................................................................57-6
C. Administration ..............................................................................................................................................57-6
D. Satellite Facilities ........................................................................................................................................57-6
III. Organizing Sensory Panels ..................................................................................................................................57-6
A. Recruiting Subjects ......................................................................................................................................57-6
B. Screening and Qualifying Subjects ..............................................................................................................57-6
C. Scheduling Subjects ....................................................................................................................................57-7
D. Personal Requirements ................................................................................................................................57-7
E. Test Orientation and Instructions ................................................................................................................57-7
F. Incentives and Motivation ............................................................................................................................57-7
G. Performance Records ..................................................................................................................................57-7
IV. Applications ........................................................................................................................................................57-7
A. Product Development ..................................................................................................................................57-7
1. New Product Development ..................................................................................................................57-7
2. Pilot Plant Scale-Up ..............................................................................................................................57-8
3. Cost Reduction ......................................................................................................................................57-8
4. Ingredient/Process Change ....................................................................................................................57-8
5. Ingredients/Purchase Specifications ....................................................................................................57-8
B. Product Quality ............................................................................................................................................57-8
1. Sensory Specification ............................................................................................................................57-8
2. Production Benchmarking ....................................................................................................................57-9
3. Manufacture Quality ............................................................................................................................57-9
4. Shelf-Life and Stability ........................................................................................................................57-9
5. Distribution Product ..............................................................................................................................57-9
C. Marketing ....................................................................................................................................................57-9
1. Monitor Competition ............................................................................................................................57-9
2. Advertising/Claim Support ..................................................................................................................57-9
3. Category Review ..................................................................................................................................57-9
4. Product Optimization ............................................................................................................................57-9
V. Discrimination Methods ....................................................................................................................................57-10
A. General Requirements ................................................................................................................................57-10
1. Subject Selection ................................................................................................................................57-10

57-1

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-2

57-2 Handbook of Food Science, Technology, and Engineering, Volume 2

2. Test Design, N, and Replication ........................................................................................................57-10


3. Data Analysis ....................................................................................................................................57-10
B. Non-Directional Discrimination Tests ......................................................................................................57-10
1. A-Not-A Sequential Presentation ......................................................................................................57-10
2. A-Not-A Simultaneous Presentation ................................................................................................57-10
3. Duo-Trio ............................................................................................................................................57-11
4. Triangle ..............................................................................................................................................57-11
5. Matching and Sorting Tests ..............................................................................................................57-11
C. Directional Discrimination Tests ..............................................................................................................57-11
1. Paired Difference ..............................................................................................................................57-11
2. n-Alternative Forced-Choice..............................................................................................................57-11
VI. Descriptive Methods..........................................................................................................................................57-11
A. Flavor Profile ............................................................................................................................................57-12
B. Texture Profile ..........................................................................................................................................57-12
C. Quantitative Descriptive Analysis (QDA) ................................................................................................57-13
1. Subjects ..............................................................................................................................................57-13
Downloaded by [University of Newcastle] at 11:02 31 October 2015

2. Screening............................................................................................................................................57-13
3. Training ..............................................................................................................................................57-14
4. Testing................................................................................................................................................57-14
5. Data Analysis ....................................................................................................................................57-14
D. Sensory Spectrum Analysis ......................................................................................................................57-14
E. Free-Choice Profiling................................................................................................................................57-14
F. Other Methods ..........................................................................................................................................57-15
VII. Time-Intensity Scaling......................................................................................................................................57-15
A. Measurement ............................................................................................................................................57-15
B. Analysis ....................................................................................................................................................57-16
VIII. Data Analysis ....................................................................................................................................................57-16
A. Descriptive Statistics ................................................................................................................................57-16
1. Central Tendency ..............................................................................................................................57-16
2. Dispersion ..........................................................................................................................................57-17
3. Frequency Distributions ....................................................................................................................57-17
B. Inferential Statistics ..................................................................................................................................57-17
1. Non-Parametric ..................................................................................................................................57-17
2. Parametric ..........................................................................................................................................57-18
C. Correlation and Regression ......................................................................................................................57-18
1. Correlation ........................................................................................................................................57-18
2. Multiple Correlation (R) ....................................................................................................................57-19
3. Regression..........................................................................................................................................57-19
D. Additional Multivariate Methods..............................................................................................................57-19
1. Multivariate Analysis-of-Variance (MANOVA) ................................................................................57-19
2. Discriminate Analysis ........................................................................................................................57-19
3. Principle Components Analysis ........................................................................................................57-19
4. Factor Analysis ..................................................................................................................................57-19
5. Cluster Analysis ................................................................................................................................57-19
IX. Visual Presentation ..........................................................................................................................................57-19
A. Histograms ................................................................................................................................................57-20
B. QDA Spider Charts ..................................................................................................................................57-21
C. Sensory Maps............................................................................................................................................57-21
X. Conclusion ........................................................................................................................................................57-23
References ....................................................................................................................................................................57-24

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-3

Sensory Science: Methodology 57-3

I. INTRODUCTION AND OVERVIEW the 1960s. The original group of psychologists chose to
not participate in the move to Natick. They disseminated
Measuring the sensory properties of foods and relating these their knowledge and interest in sensory evaluation further
properties to consumer acceptance is a major objective and into the food industry by joining product companies (e.g.,
accomplishment for sensory evaluation. These achieve- Coca Cola, Pillsbury, and General Mills), contract research
ments are possible as a direct result of advances in the sci- organizations (e.g., Battelle Memorial Institute) or founding
ence, development of valid and reliable methods, education research companies (e.g., Peryam and Kroll Corporation).
of research and application professionals, development of Government and industry interest in sensory evalua-
cost efficient and well run sensory programs, and institu- tion encouraged additional developments. For example, in
tions and commercial companies that support these pro- the 1950s the Arthur D. Little Company introduced the
grams. The latter is directly related to the expected and Flavor Profile Method (discussed below), a qualitative form
realized commercial benefit of sensory tests. of descriptive analysis that did not rely on an individual
This chapter provides the reader with an overview of expert, and formalized subject screening and training pro-
sensory evaluation, its uses and limitations. We highlight cedures. Private contract research companies continue to
important developments, current thinking and the current play an important role through sponsoring workshops,
status of the science. The interested reader will find more developing proprietary methods, assisting product compa-
detail in several books published on the topic, most nota-
Downloaded by [University of Newcastle] at 11:02 31 October 2015

nies establish internal sensory capabilities, and conducting


bly, Lawless and Heymann (1), Stone and Sidel (2), and research studies.
Amerine, Pangborn, and Roessler (3). By the mid-1950s, a few universities were offering a
series of courses in sensory evaluation; most notably the
A. HISTORICAL BACKGROUND University of California at Davis (UCD). Many students
from the U.S. and abroad introduced the science into com-
The food industry, similar to other consumer product indus- panies, while others went on to teach the science at the uni-
tries, traditionally viewed sensory evaluation in the context versity level. Currently, a wide range of universities and
of the technical “experts” who through education (e.g., technical institutes worldwide offer programs in sensory
learning to judge the quality of dairy products as part of evaluation, and in 2002 UCD initiated a sensory course
dairy science education) and years of accumulated experi- online: “Applied Sensory Science and Consumer Testing
ence were able to describe company products and set stan- Certificate Program for Distance Learners.”
dards of quality by which raw materials would be purchased Also during the era of the 1950s and 1960s, various
and products manufactured and marketed. Three noteworthy technical and scientific societies organized activities and
developments led companies to evolve away from reliance published procedures focusing on sensory evaluation and
on the single technical expert, and to reliance on formal sen- the measurement of flavor. These included Committee
sory panels and procedures. First, products became more E-18 of the American Society Testing and Materials, the
complex, and technical experts were less able to predict con- Food and Agriculture Section of the American Chemical
sumer acceptance. Second, companies could not wait the Society, the European Chemoreception Organization, and
2 or more years for an “apprentice” to assume evaluation the Sensory Evaluation Division of the Institute of Food
responsibilities. Thirdly, the emerging field of sensory eval- Technologists.
uation was demonstrating the need for, and value of, formal,
unbiased, and statistically relevant procedures for qualifying
B. DEFINING SENSORY EVALUATION
subjects, evaluating products, and communicating action-
able results. The Sensory Evaluation Division of the Institute of Food
During the 1940s and 1950s, sensory evaluation Technologists defined sensory evaluation as “a scientific
received additional impetus through the U.S. Army’s Quar- discipline used to evoke, measure, analyze and interpret
termaster Food and Container Institute, which supported reactions to those characteristics of foods and materials as
research in food acceptance for the armed forces (4). It they are perceived by the senses of sight, smell, taste,
became apparent to the military that adequate nutrition, as touch and hearing” (5). This definition emphasizes several
measured by analysis of diets or preparation of elaborate important points, including 1) the scientific focus of the
menus, did not guarantee food acceptance and consump- discipline, correctly distancing the discipline from art and
tion by military personnel. The importance of the sensory opinion; 2) although implied and not directly stated, the
components of food was acknowledged, and resources foods and materials represent physical, rather than cogita-
allocated to studies for determining how best to measure tive, stimuli; 3) implicit in the definition is the concept of
food acceptability. Several psychologists were hired by the the measurement and analysis of perceptions from infor-
military to develop methods for measuring food accept- mation obtained through the five senses; and 4) the defi-
ance, and this practice was continued when the Army’s nition forwards a distinction between the perceived
food research laboratories were moved to Natick, MA in characteristics of foods and materials and their actual

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-4

57-4 Handbook of Food Science, Technology, and Engineering, Volume 2

physical properties. To this latter point, the definition where mechanical devices for measuring texture have been
focuses on the importance of perception for understanding described in the literature for more than a century. Bourne
the impact of a product’s physical properties; implicit in this (6) provided an excellent review of the subject describing
is the notion that in sensory evaluation “perception is real- mechanical devices to measure product changes, particu-
ity.” This has enormous implications on determining how larly the fruit ripening process, or to simulate the chewing
subjects are selected, and, when necessary, trained, what process to measure meat tenderness. The interested reader
sensory evaluation should measure, and the measurement is also referred to the Journal of Texture Studies, especially
method. the publications of A.S. Szczesniak and her collaborators.
The evolution of sensory evaluation as an independent While such mechanical devices have undergone substan-
discipline in the 1940s and 1950s produced unique meth- tive change in terms of sensitivity, measurement, design,
ods for determining difference. These and other sensory etc., the basic principles have remained unchanged, to wit,
method developments are a reminder that although they creating a mechanical device to measure what humans
have a common lineage, psychophysics and sensory eval- perceive.
uation have very different objectives. Sensory evaluation The main difficulty in simulating sensory behavior by
focuses on measuring the responses of people to products, an instrument is due to the complex simultaneous activity
and application of findings to action-oriented decisions ongoing in the mouth during eating and drinking. The upper
Downloaded by [University of Newcastle] at 11:02 31 October 2015

about products. Psychophysics focuses on the measure- jaw remains stationary while the lower jaw is capable of sev-
ment process and explaining the behavioral mechanisms eral different motions to hold, bite, tear, and masticate. At
behind sensory responses. Both objectives are important, the same time, the highly muscled tongue gathers, moves,
and each has its own criteria for success. and compresses stimuli, the salivary glands contribute mois-
ture and lubricant to the process, and swallowing provides
control of volume in the mouth while momentarily limit-
C. CLASSIFICATION OF SENSORY METHODS
ing intake to the mouth. To our knowledge, this multi-
There are several sensory methods, differentiated primarily dimensional activity cannot be reproduced with current
by objective, measurement procedure, subject qualification, instrumentation.
and various combinations of these factors; and new methods The history of odor detection has always included both
continue to be developed. As reported in Stone and Sidel (2), human and instrument systems. The earliest instruments
sensory tests are classified primarily by type of test in a three- began with measuring total volatiles. This was followed by
category classification system, which are: discriminative, development of chromatographic techniques and gas-liquid
descriptive, and affective tests. (We view discriminative and chromatography (GLC) which enabled separation of chem-
descriptive as analytical tests and thus could consider two icals as well as ability to separate specific chemicals at
categories; analytical and affective tests.) There are several very low concentrations; e.g., ppm (parts per million), ppb
different tests within each class, and they are described in (parts per billion) and less, to facilitate their identification.
greater detail later. Discrimination tests measure whether Subsequent developments enabled integration of the chro-
there is a perceived difference between products, descriptive matographic separation with even more sophisticated
tests measure perceived attributes and their intensities for instrumentation such as mass spectroscopy to more pre-
a product, and affective tests measure product acceptance or cisely clarify chemical structure, and so forth. Amerine et
preference. Affective tests are discussed in the current al. (3) describe limitations and early efforts to combine
handbook under the topic of Consumer Testing. GLC with sensory panel results. Volatiles from foods are
trapped and collected for sensory evaluation. Later refine-
ments included development of equipment containing mul-
D. SENSORY VS. INSTRUMENTAL TESTING
tiple odor ports that delivered the separated odorant at
The development of instruments to measure food and bev- prescribed dilutions simultaneously to sensory panel mem-
erage characteristics such as appearance, aroma, taste, and bers. Gas chromatography-mass spectrometry are used
texture has a very long history, as has research into corre- to analyze perceived odors, and Noble (7) described the
lating instrumental and sensory measures. Interest in this use of instrumental analysis for the sensory properties of
relationship in the food and beverage industry is based in foods.
part because it represents a potential means of reducing During this same time period, Schutz, Veley, and Iden
reliance on the human judgment, which some view as vari- (8) and Wilkens and Hartman (9) described their efforts to
able and subjective. Value, efficiency, and economy are devise other systems for the detection of odorants. Schutz
possible where instrumental measures can be effectively and co-workers investigated use of specific enzyme sys-
substituted for sensory measures. This is especially true in tems associated with olfactory tissue that were reactive
product quality assessment. to odorants and correlated results with the physical and
An extensive literature is available describing sen- chemical properties of those odorants. At the time, the con-
sory/instrumental relationships for texture measurement, straints of separation technology, enzyme stability, and

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-5

Sensory Science: Methodology 57-5

related issues impeded further development of this A. TESTING


approach. Wilkens and Hartman, on the other hand, took
a very different approach to the problem; they devised an A product evaluation area typically consists of the following
instrument, an electronic analog that relied on microelec- three functional areas, adjacent to one another and sepa-
trodes made from combinations of metallic wires for the rated by walls and doors for easy communication and pas-
detection of odorants. These authors were able to demon- sage. This area is best located for its ease of accessibility to
strate differential response sensitivity to various volatile the subjects and for its low level of odors, noise, and vibra-
chemicals with just a few microelectrode metal combina- tion associated with more intense R&D and manufacturing
tions. Today’s “Electronic Noses” and “Electronic Tongues” activities.
represent significant advances in electrode specificity,
sensitivity to individual chemicals, use of computer chip 1. Reception Area
technology for analyses, and in accessing pattern recogni- The reception area is typically a well-lit and comfortable
tion and neural network statistics as an aid in data inter- room located off a main corridor, and offering easy access
pretation (10, 11, 12, and 13). Bleibaum et al. (11) to the test area. It is equipped with a receptionist desk and
concluded from their research “the electronic tongue and chair, telephone for contacting subjects, comfortable chairs
electronic nose, in combination, can be used to predict the where subjects are either seated while receiving instruc-
Downloaded by [University of Newcastle] at 11:02 31 October 2015

sensory characteristics and their relationship to the qual- tions or where they wait prior to proceeding to an assigned
ity of apple juices measured by consumers.” The degree to sensory booth. A blackboard or bulletin board is located
which these odor and taste detection instruments, or those on a wall near the receptionist desk and identifies the prod-
yet to be developed, are incorporated into routine product uct category and number of products that are to be tested
development and quality control activities will depend on in the session. A table is also provided for those situations
how well they correlate to important sensory measures, where subjects are offered a reward, or “treat” (juice, cof-
and are economical and safe to operate in those respective fee, cookies, candy, etc.), immediately following participa-
environments. tion in a test. The room should allow easy access to, yet
Correlating sensory flavor attributes and chemical visual separation from, the sensory booth area to minimize
measures has become almost routine in research and qual- distraction from early arriving subjects to subjects still
ity control, as has correlating sensory and Hunter color seated in the sensory booths. The dimensions for the recep-
measures. Instrumental and sensory correlations also have tion area depend on the number of subjects it has to accom-
been described for visual clarity and turbidity (14). modate prior to their proceeding to the sensory booths.
Before leaving this topic, we are reminded that sensory
information is uniquely different than information obtained
2. Test Kitchen Area
from a physical instrument (a point often overlooked when
creating a mechanical system to substitute for a physiolog- The test kitchen area and materials used in this area typi-
ical system). It represents the relatively simultaneous inte- cally represent a well-equipped commercial, or heavy duty
gration of information from all the senses, in conjunction home, kitchen. The area must accommodate preparation
with cognitive factors that include learning and expectation. and cleanup equipment, equipment and supply storage,
In this sense, sensory information is multidimensional product preparation, holding and serving activities, and
whereas instrumental information is almost entirely one workstations for kitchen staff. One or two large central
dimensional. Although we can instruct subjects to not pay islands for product preparation, showings, and holding for
attention to certain attributes, it is false to assume that those serving are recommended, and these must allow for unen-
attributes do not influence the subject or his response. cumbered foot traffic flow around them during product
Instruments can be useful in sensory evaluation when there preparation and serving. Airflow from this area is an impor-
is a good understanding of the relationship between those tant consideration, and food preparation aromas from here
measures, sensory perception, and consumer behavior. At must be exhausted away from the sensory booths. Serving
present such relationships are the exception, and consid- counters need to be high enough to minimize posture
erably more needs to be accomplished before instruments stress during serving, and wide enough to accommodate
represent a practical alternative to sensory measures. products, supplies, and scorecards for subjects seated in sen-
sory booths with pass through into in the serving area. All
under- and above-counter space should be utilized for stor-
II. SENSORY EVALUATION FACILITIES
age, and an adjacent storage room is recommended where
The typical sensory evaluation facility in a R&D center possible. A door from the kitchen to the reception area will
consists of separate areas for testing, panel training, and facilitate any needed communication between the recep-
administration. Satellite facilities are sometimes required tionist and kitchen staff, and provide easy access to treats
to accommodate testing that cannot be accomplished in and materials needed to clean and maintain the reception
the immediate vicinity of the research center. area. Telephones in this area should be equipped with a

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-6

57-6 Handbook of Food Science, Technology, and Engineering, Volume 2

visual alert, so that the auditory alert may be turned off rules for maintaining these locations apply; that is, a clean,
during testing. quiet, and distraction-free test environment.

3. Sensory Booths
III. ORGANIZING SENSORY PANELS
Individual sensory booths need to minimize visual and other
There are several steps to developing and organizing effec-
potential distractions to the subjects from one another, from
tive sensory panels. Subjects must be recruited and quali-
the kitchen, and from the reception area during the evalu-
fied, must participate in tests, be motivated, and have their
ation process. Dimensions, surface colors and materials,
performance monitored.
lighting, ventilation, pass-through structure, expectoration
system, signal system, and direct data entry systems must
be considered when designing sensory booths for food A. RECRUITING SUBJECTS
research. Clean filtered air should flow into the booth area, Subjects for sensory panels may be recruited from a gen-
and air should flow from the booths to the kitchen, and not eral population of consumers local to the sensory facility,
the reverse. The interested reader will find useful booth or from a company’s employee list. Because sensory panel
design information in several texts, and is directed to membership is performance related, the decision about
Eggert and Zook (15) and Stone and Sidel (2) for additional
Downloaded by [University of Newcastle] at 11:02 31 October 2015

which group is best to recruit is influenced by factors of con-


detail. Most sensory booths are permanent structures; how- venience, cost and availability. Subjects may be recruited
ever, temporary booths can be constructed from sturdy randomly by telephone, mall intercept, or electronically.
coated art board or plywood, and these can be made col- Alternatively, recruitment notices can be sent to prospective
lapsible or foldable with hinges or fitted slots for portability. subjects, or posted in locations frequented by them. The
Temporary booths can be moved to off site testing locations. recruitment notice should state that people interested in
joining a food evaluation panel are being sought, and
B. PANEL TRAINING individuals who qualify will be required to participate for
1 to 3 hours a day for an extended period of time. People
Open discussion is required at various times during panel
responding positively to the recruitment request are
training, and these discussions can be held in a conference
screened further for their availability and willingness to
room, focus group facility, or a dedicated training room. The
evaluate specific food categories that will be tested. To
room should be proximal to the test kitchen or product
reduce early panel attrition, it is advisable that subjects are
preparation area, be readily accessible by subjects and sen-
likers and users of foods in the category. Twenty-five to
sory staff, and free from distraction as described above.
thirty interested and qualified subjects should be invited
Sufficient blackboard, wall, or easel space is required to
for sensory screening.
accommodate attribute lists developed by the panel during
training. A one-way mirror to an adjacent viewing room,
equipped with video and audio taping capabilities, can B. SCREENING AND QUALIFYING SUBJECTS
accommodate observers without distracting subjects during Sensory panels require subjects who are qualified based
training. on their sensory skills and availability. A variety of tests
are used to determine sensory skills, with some being
C. ADMINISTRATION method specific (e.g., Texture Profile®). Most popular, but
The staff required to support project planning, report writ- not most useful, screening methods used by sensory pro-
ing, data processing, and general administrative functions fessionals are: basic taste and odor thresholds, basic taste
should be located in a common area accessible from a intensity ranking, identification of selected taste and odor
main corridor and near to the test kitchen. Access to the compounds, texture stimuli for texture-specific panels, and
administrative area should not require passage through the discrimination of simple compounds and more complex
test area. finished products. Successful screening, especially when
based on the latter procedure, can eliminate poor per-
forming subjects, but does not guarantee successful per-
D. SATELLITE FACILITIES
formance. No evidence has been provided that demonstrates
Most sensory tests are conducted in well equipped labora- that threshold testing or use of “standards” actually iden-
tories as described above and are associated with company tifies better performing subjects. However, eliminating
headquarters and a main research center. Tests also may be subjects who cannot discriminate product difference at bet-
conducted in locations specific to product quality mainte- ter than chance has proven to be a good predictor and is suf-
nance and production. Such satellite facilities are smaller, ficient justification for continuing the latter practice. For
as they require less preparation equipment and space, and discrimination and descriptive panels we recommend using
fewer booths. Regardless of where done, the same general discrimination screening tests within the test product

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-7

Sensory Science: Methodology 57-7

category. Screening should involve at least ten product Positive motivation begins with informing subjects that
pairs, given in replicate, covering all sensory modalities the task is important, maintaining an efficient and profes-
and increasing in difficulty from easy (85 to 100% success sional test program, and providing positive reinforcement
rate) to very difficult (50% or less success rate). where possible, without disclosing information that can
influence upcoming tests. Following testing, a daily treat,
C. SCHEDULING SUBJECTS such as a beverage, baked good, confectionery, fruit, veg-
etable and dip immediately reinforce subject participation.
Subjects should be scheduled to arrive at the sensory facility These items need to be thoughtfully presented and fresh.
at the same time, receive their instructions as a group, pro- Once they become routine or are leftovers from previous
ceed to their assigned booth, and begin their evaluation. tests, they lose their reinforcing value. Where there is insuf-
Scheduling avoids conflicts, assures similar instructions and ficient space for a treats table, or where one would interfere
products are given, provides order, and establishes this as a with testing, a coupon of equal value and redeemable at the
serious professional activity. Allowing subjects to arrive at company cafeteria can be substituted. Holidays provide the
the facility and begin testing whenever they like introduces opportunity to incorporate a special theme into the daily
potential error from a variety of sources including product treat.
stability, extending the session too long, and traffic conges- Long term incentives include various forms of recogni-
tion in the test area. Most testing is done in the morning
Downloaded by [University of Newcastle] at 11:02 31 October 2015

tion rewarding participation and attendance rather than skill.


from 9:00 to 11:30 and in the afternoon from 1:30 to 4:00. A letter of commendation from a senior executive, com-
Suppliers that use consumers trained for sensory panels may pany newsletter recognition for the “taster of the month,”
schedule testing for the evening hours until 9:00 or 9:30 pm. an annual dinner out or panel party, and accumulating points
toward a gift or raffle have proved successful. Incentives that
D. PERSONAL REQUIREMENTS reward the subject and publicize the sensory program are
always a good idea. Cash payments are usually reserved for
Subjects should be informed that they are expected to
non-employee subjects. Employees should not be compen-
arrive on time at the test location, and to not have con-
sated for their participation.
sumed food or flavored beverages, or smoked for approx-
imately ½ hour before testing. They are asked to not use
fragranced personal care products, perfumes, and so forth, G. PERFORMANCE RECORDS
when reporting to the test site, and they are cautioned to Sensory panels use a small number of subjects; therefore,
not wear clothing that has retained aromas from previous the performance of each subject is important. Performance
contact with these products. Subjects are discouraged from records track the number and type of panels on which the
discussing product tests other than during sessions con- subject participates and notes any sub-standard areas of per-
ducted by the panel leader. In many cases they are required formance. An individual subject’s failure to find differences
to sign confidentiality agreements, and in tests involving where the panel result is statistically significant is noted in
experimental ingredients or processes may be asked to the performance record. The information will help to iden-
sign release of liability forms. tify where additional training is required for the subject or
the panel, or when a subject needs to be excused from a
E. TEST ORIENTATION AND INSTRUCTIONS panel.
Subjects reporting to a site are provided with a verbal ori-
entation. The orientation informs subjects about the prod- IV. APPLICATIONS
uct category, number of products to be evaluated, the The growth and success of sensory evaluation is directly
evaluation procedure, and any procedural variations that related to its positive contributions and variety of applica-
differ from previous evaluations. Written instructions are tions to product development and product quality. The fol-
posted in each test booth, and subjects are encouraged to lowing section describes many of these useful applications.
read those instructions before evaluating the first product,
and again during rest intervals between products.
A. PRODUCT DEVELOPMENT
F. INCENTIVES AND MOTIVATION Product development is an activity that ranges from new
product development to several planned activities associ-
Positively motivated subjects are more likely to follow ated with changes to a product or process.
test rules, have lower panel absenteeism, remain focused
on the task, and are less likely to be negative and disrup-
1. New Product Development
tive. Although positive motivation does not guarantee sat-
isfactory performance, it increases that likelihood, and is The new product development cycle begins with a prod-
beneficial for the other subjects and for the test staff. uct idea generated from consumer research or technical

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-8

57-8 Handbook of Food Science, Technology, and Engineering, Volume 2

discovery, and is complete when marketing research current product. A difference usually results in rejection
demonstrates sufficient consumer interest in the new prod- of the cost-reduced product. Where a difference is found,
uct. The amount and detail of direction consumer research descriptive tests may describe differences to determine
will provide about the sensory characteristics for a new whether the difference is for an important attribute or one
product will differ based on a company’s ability to obtain that may be readily corrected. Caution is advisable rela-
that information. Trained panel (e.g., QDA) evaluations in tive to using a series of discrimination tests where a suc-
these consumer studies will provide an accurate description cessful cost-reduced product replaces the control. For
of product differences and similarities, and when combined example, if A is not significantly different compared to B
with consumer preferences will identify sensory attributes and B is not significantly different compared to C, and so
and intensities to optimize consumer acceptance. The sen- forth, A very well may be significantly different when
sory description for the optimized product becomes the compared to Z. This transitivity can occur when a series
development target. Trained descriptive panels and quali- of non-significant differences result in a product that is
fied discrimination panels are used to determine when one significantly different from the original control. This pos-
or more products satisfactorily achieve or approximate a sibility can be minimized through the use of descriptive
sensory optimized target. analysis, comparing the descriptive profile of the proposed
new control to that of reproduced original control or to the
Downloaded by [University of Newcastle] at 11:02 31 October 2015

historical profile for the original control.


2. Pilot Plant Scale-Up
A pilot plant is often used during product development as 4. Ingredient/Process Change
a means for providing test products and quantities sup-
porting research and development without interrupting Alternate ingredients, new equipment, and process changes
manufacturing or requiring production size quantities. At may be introduced at any time. All changes should be
the conclusion of the R&D development process, the pilot evaluated to determine their potential effect on product per-
test product (i.e., formulation and process) will need to ception; the evaluation strategy is the same as described
be converted to a production product, using production for cost reduction. In situations where a change is required
equipment and processes. Differences between pilot plant by the business, even where sensory differences are
and production product usually is unwanted; however, found, consumer tests are recommended to estimate the
they are not unexpected. The manufacturer needs to deter- impact of the change.
mine that there is a perceived difference between the two
products, and if there is, what are the implications of that 5. Ingredients/Purchase Specifications
difference. Where possible, sensory discrimination tests
are used to determine whether there is a perceived differ- Sensory requirements are designed into the purchase spec-
ence between the pilot and production products, and where ification for many ingredients. For example, a prescribed
a difference is found, trained panel descriptive tests are perceived heat level is required for hot peppers that will be
used to describe those differences. Product Development used in processed salsa.
will need to determine formulation or process variables
responsible for the difference and make necessary changes. B. PRODUCT QUALITY
If the production product is different compared to the pilot The objective is to provide a consistent product of appro-
plant product and cannot be modified further, it’s impact priate sensory quality, where appropriate sensory quality
on acceptance will need to be assessed. This is achieved is defined during the product development cycle. The fol-
either through comparison to an optimization model, if lowing describes important areas where sensory evalua-
one was developed during the development process, or in a tion supports the quality mission.
consumer test of the pilot and production product. Where
production product cannot be modified further and it does
1. Sensory Specification
not satisfy the company’s action standard for acceptance
for new product introduction, the new product may be A sensory specification describes the acceptable range for
abandoned. the most important characteristics of a product. Specifica-
tion development requires products selected to represent
the broad range of production product and possible varia-
3. Cost Reduction
tion. Products are evaluated by a trained panel and by con-
Cost reduction is an ongoing activity, and where possible sumers. The trained panel provides sensory descriptions
sensory discrimination tests are used to determine (e.g., QDA) for each product, and consumers, qualified as
whether there is a perceived difference between the con- likers and users of the product, provide the liking scores.
trol and cost-reduced product. Where no difference is The data are analyzed to identify the most important accept-
found, the cost-reduced product is accepted to replace the ance predictors (usually about 10 to 12) and the tolerance

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-9

Sensory Science: Methodology 57-9

range for each, and these results are used to train quality continuing through tracking product performance, and
control panels at manufacturing sites. contributing to special assignments such as developing tests
and data to support or challenge advertising claims. We
2. Production Benchmarking briefly describe the role of sensory evaluation in those activ-
ities.
Once a new product has been accepted for manufacture
and distribution, production samples are evaluated by a
1. Monitor Competition
trained descriptive panel to establish the sensory profile
for the product when first released for sale. The bench- Discrimination and descriptive panel information are used
marking documentation is used for future comparisons to provide immediate intelligence about the similarity and
and references to the original product. differences between competitive products and also for
tracking changes and trends over time.
3. Manufacture Quality
2. Advertising/Claim Support
In a manufacturing facility, subjects are screened and trained
to evaluate a product at one or more stages of production Sensory claims often appear on television, radio, and in
(and where permitted according to GMP and HACCP mass media print and trade oriented publications. Most
Downloaded by [University of Newcastle] at 11:02 31 October 2015

requirements). These evaluations can include ingredient claims are comparative and preference related, others may
receipt and storage, during manufacture, and finished prod- identify specific sensory differences. In either case, the
uct. For finished product, trained sensory panels evaluate claims can be challenged and damages assessed where the
attributes included in the sensory specification. A sensory claim is not substantiated. All phases of the research,
professional coordinates the activity but is not a panel including subject recruitment, selection, training, testing
participant. Subjects evaluate attribute intensities, and the and the data analysis are subject to examination and chal-
sensory professional uses that information to determine lenge. Therefore, it is important that sensory claims are
whether to recommend accept, hold for further evaluation, supported by appropriate research that can be defended in
or reject the product. the legal system. The research must demonstrate that it is
free of bias that could influence the outcome, and that sound
testing principals are used.
4. Shelf-Life and Stability
The sensory quality of products deteriorates over time, and 3. Category Review
the rate of deterioration depends on factors related to ingre-
dients, packaging, handling, and storage. Several different A broad array of products, often between 12 to 24 and
sensory related criteria are possible for determining when representing the competitive and sensory range for a cate-
the end of shelf-life occurs, and no one criterion is univer- gory, are evaluated by a trained descriptive panel and by
sally accepted. Using a qualified descriptive or discrimi- consumers. Reviews for a rapidly changing product cate-
nation panel, any statistically significant difference for a gory may be scheduled as frequently as 6 month or 12
sensory attribute compared to a control product is a popular month intervals. The information is used to monitor the
choice for determining end of shelf-life, albeit a conserva- competitive market on a broad scale and to track changes
tive one. Where a sensory specification is available, stored and trends over time.
product is evaluated to access whether it deviates from
that specification. A loss of consumer acceptance is another 4. Product Optimization
popular criterion used to determine end of shelf-life.
This category of sensory research is conducted for the
purpose of optimizing consumer liking for a product. Two
5. Distribution Product different approaches are used for selecting products for
Products may experience abuse after leaving the manu- this research; one is a design of experiments (DOE), and
facturing facility; therefore, many companies audit product the other is a category review (CR). The DOE limits prod-
quality at the retail level. Results from qualified descrip- ucts to those the researcher modifies according to a pre-
tive and discrimination sensory panels are used to measure scribed experimental design (e.g., factorial, RSM); the CR
differences compared to data from the product benchmark includes a range of products having sensory differences
or sensory specification. representing the product category. A qualified discrimina-
tion panel may be used to eliminate redundant products
and a trained sensory panel to provide a descriptive analy-
C. MARKETING
sis of the products. The statistical analysis relates the con-
Sensory evaluation supports marketing and marketing sumer, trained panel, and analytical (if included) data sets
research activities beginning with new product development, to determine combinations of sensory and analytical

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-10

57-10 Handbook of Food Science, Technology, and Engineering, Volume 2

attributes that best predict optimal consumer liking. The characteristics of a control product that is a benchmark. For
descriptive panel is used for the follow-up research to eval- each trial in this fixed order scenario, two of the three prod-
uate test products until one or more satisfy the optimized ucts are the control and the third is the test product.
sensory target. Sensory analytical tests typically use small numbers of
subjects screened for their sensory skills, and the discrimi-
nation model is no exception. About 12 screened and qual-
V. DISCRIMINATION METHODS
ified subjects are optimal, and together with a recommended
Discrimination tests are classified as analytical methods, replicate trial provide sufficient data for statistical analysis.
and are used primarily to determine whether there is a per- Replication is recommended for analytical sensory tests,
ceived difference between products. The tests may be because each subject contributes a large percent of the data
directional, where the attribute of difference is named, or and it is important to know which subjects have difficulty
non-directional, where the attribute of difference is not replicating their response. An inability of a subject or a panel
named. Discrimination tests are relatively uncomplicated to replicate responses can provide important information
to administer (most use a simultaneous sample presenta- about problems with the subjects, procedure, and products.
tion), analyze, interpret, and report. They are used in sensory
studies primarily to screen products as a prelude to other 3. Data Analysis
Downloaded by [University of Newcastle] at 11:02 31 October 2015

types of tests (e.g., descriptive or consumer), or to screen


subjects for participation in descriptive and other sensory Unless noted otherwise below, statistical testing for dis-
analytical panels. Discrimination tests are used in quality crimination data is typically based on expansion of the
control and manufacturing to identify non-conforming prod- Binomial distribution. Published tables are available in
uct. They may not be the method of choice where inherent the literature for determining the number of correct judg-
variability within samples is large. ments required for statistical significance (2, 15). These
Discrimination testing dates back to the psycho- tables provide for 1-tail and 2-tail tests and for chance
physics of the 1800s and procedures used for absolute and probability values of 0.5 and 0.33.
difference threshold determination. The following section
discusses discrimination tests used by sensory scientists B. NON-DIRECTIONAL DISCRIMINATION TESTS
to make product decisions, without reference to competing
The following tests are classified as non-directional
psychophysical theories or models attempting to explain
because subjects are not instructed to select the different
the discrimination process.
product based on any specified attribute.

A. GENERAL REQUIREMENTS
1. A-Not-A Sequential Presentation
Following are some requirements for discrimination tests.
We are reminded that these tests are sensory analytical Subjects are provided with two 3-digit coded products,
tests, and as such use small numbers of screened and qual- one followed by the other, and indicate whether the sec-
ified subjects. By small numbers, we mean 20 to 25; how- ond product is the same, or different from the first. The
ever, there is no rule that precludes more or fewer subjects. first product is removed prior to serving the second to
eliminate a direct comparison of the two. This method is
used primarily to minimize observation of unimportant
1. Subject Selection visual differences between products, and to control expo-
Subjects for sensory analytical tests are screened for their sure time between products having significant carryover
sensory skill, and for discrimination tests this means for effects. The probability of a correct judgement is 0.50;
their ability to detect differences between products. Using data analysis is straightforward and based on the binomial
consumers or other subjects that have not been screened expansion, for which there are several published tables.
for sensory acuity is not common practice or is it recom- The A not-A method can be expanded to where two or
mended. Unskilled subjects increase the risk of a Type II more coded products are given to the subject, one at a
error (i.e., missing a real difference); large numbers of sub- time, without reintroduction of the first product. Although
jects, typical for consumer tests, increases the risk of a Type this further reduces carryover effects, selection of this
I error (i.e., finding a false difference). A previous section serving option must be balanced against unwanted differ-
provides a detailed discussion about subject screening. ences in subject’s memory skills.

2. Test Design, N, and Replication 2. A-Not-A Simultaneous Presentation


Balanced design serving orders are typical for discrimina- Subjects are provided with 2 products, side by side, and
tion tests within and between product trials, except for those indicate whether the products are the same or different.
three sample tests where it is prudent to reinforce the sensory They are instructed to try the product on the left first, and

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-11

Sensory Science: Methodology 57-11

then the product on the right. This design should include They are instructed to try the product on the left first, and
an equal number of pairs in which the products are in fact then the product on the right. The order of tasting is bal-
the same, to reduce the impact of false hits from subjects anced across subjects and replications.
that always respond that there is a difference. The primary criticisms of this method, and of all direc-
tional discrimination tests, is the time and effort required
3. Duo-Trio to train subjects to evaluate the specific attribute, and know-
ing whether subjects actually selected a product based on
Three products are presented simultaneously to the subject,
the specified criteria. Commercial products are often quite
where one is coded “R,” for reference, and the other two
complex, where an ingredient or process difference can
products have unique 3-digit codes. The subject is instructed
result in several sensory differences. Limiting the subject’s
to select which coded product is most similar to the refer-
response to a single attribute may result in a correct choice
ence. Similar to the A-not-A method, it does not require
based on incorrect criteria. For example, different sweet-
selection based on a named attribute or training with that
ener types and amounts for a beverage or baked product
attribute, has a p ⫽ 0.50, and one can use published tables
may alter visual and textural characteristics. Products scored
for the binomial expansion. As we will see, this method
different for sweetness actually may differ on some other
requires less re-tasting and consequent potential for carry-
attribute. Several investigators (1,18,19,20) have convinc-
over effects than does the Triangle method. A sequential
Downloaded by [University of Newcastle] at 11:02 31 October 2015

ingly argued that, all things being equal, a directional


procedure can be used to further reduce direct compar-
method is superior to an equivalent non-directional method.
isons and carryover effects; however, it introduces unwanted
The problem is, seldom are all things equal. Where attrib-
memory skill issues.
ute information is required, sensory descriptive methods
are recommended.
4. Triangle
Three 3-digit coded products are presented simultane- 2. n-Alternative Forced-Choice
ously, and the subject is informed that two of the products
The directional paired difference test can be expanded to
are the same and one is different. The subject’s task is to
include several samples, and the subject instructed to select
correctly select the product most different from the other
the product that has the most, or least, of some attribute.
two. The triangle method requires more tasting to reach a
As in all discrimination tests, the probability value must
decision than does the duo-trio method. This is because
reflect the number of samples from which the subject must
the triangle has 3 pairs of unknown products (A vs. B, A vs.
select. For example, the probability value is 0.33 for a
C, and B vs. C) compared to two pairs (Reference vs. A,
3-product test, 0.20 for a 5-product test, and so forth. AFC
and Reference vs. B) for the duo-trio. The chance probabil-
tests are directional and have the same procedural problems
ity for the triangle test is 0.33 to reflect that there are three
described for the paired difference test. They are multi-
unknowns, and there are published tables available for
sample tests where all the samples for a trial are simulta-
assessing significance between products.
neously presented, introducing issues of sensory fatigue
for many commercial products.
5. Matching and Sorting Tests AFC methods are have been popularized most recently
A variety of non-directional discrimination tests are pos- by researchers developing theoretical models to explain
sible for matching unknowns to two or more controls or sensory discrimination behavior (19,20). Signal Detection,
standards (e.g., dual-standard method), sorting unknowns Thurstonian, and other competing approaches may well pro-
into similarity groups, or indicating degree of “sureness” vide answers to important sensory discrimination issues,
that an unknown matches a previous sample. An example and their examination is encouraged. To be useful for rou-
of the latter is the R-index described by O’Mahony (17). tine product evaluation, these approaches will need to
These methods require more tasting than do the methods offer alternate procedures that are practical in a business
described above, and for this reason are not routinely used environment.
for product evaluation.

C. DIRECTIONAL DISCRIMINATION TESTS VI. DESCRIPTIVE METHODS


Descriptive analysis is the most sophisticated source of
For directional difference tests, subjects are instructed to
information available about the perceived sensory proper-
make their choice based on a specified attribute.
ties of products. It is particularly important because it pro-
vides a focus for development efforts. Descriptive analysis
1. Paired Difference
provides a basis for measuring the effects on perception
Subjects are provided with 2 products, side by side, and of a process or of ingredients; it is used for correlating
indicate which product has more of a specified attribute. instrumental measures, and it is essential for correlating

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-12

57-12 Handbook of Food Science, Technology, and Engineering, Volume 2

consumer response behavior and for identifying those for selected chemical compounds, and a personal inter-
product attributes that are most important to consumer view to evaluate interest, attitude and availability. Training
preferences. Before discussing specific methods, it is nec- takes 6 months or more and includes instructional infor-
essary to define what is meant here by descriptive analy- mation about the senses, introduction to the components
sis. Descriptive analysis is a sensory methodology that of the flavor profile, and direct experience evaluating
provides quantitative descriptions of products based on selected products. Training also includes threshold deter-
the perceptions from a group of qualified subjects. Based mination for the basic tastes, skills having little or no con-
on the method, it provides information about a specific nection with evaluation of products containing mixtures
sensory modality or provides a total sensory description, of supra-threshold stimuli. Selected references are intro-
taking into account all the sensations that are perceived, duced to demonstrate specific sensations, and practice
visual, auditory, gustatory, olfactory, and kinesthetic — profiles obtained for products evaluated by an experi-
when the product is evaluated. The evaluation is usually enced panel. During training all subjects serve as panel
(but not always) done in a controlled environment, and leader until one is selected. The panel leader is key to this
can include product handling, and, in this sense, it is a total method; they coordinate the testing and report results. This
experience. The evaluation also can be done in a home individual assumes a leadership role, directing the conver-
environment when normal preparation, use, and consump- sation and providing a consensus conclusion based on the
Downloaded by [University of Newcastle] at 11:02 31 October 2015

tion are required over a period of time. results. The role of panel leader as defined in the method
The origins of descriptive information can be traced to can have significant consequences without some inde-
early brewmasters, perfumers, flavorists, and other product pendent controls. Subjects could be led to a conclusion
specialists. These technical specialists described products, without being aware this had occurred.
made recommendations about the purchase of specific raw The method had considerable appeal because results
materials, and also evaluated the effect of process variables could be obtained rapidly. Subjects meet, as a group, for
on product quality (as they determined product quality). about an hour to evaluate a product, reach a consensus about
Companies used that information to determine that a par- its sensory properties, and provide the requestor with a
ticular product met their criteria for manufacture and sale to result. The developers of the method emphasized that there
the consumer. These early activities were the basis for the would be confidence based on the collective professional
foundation of sensory evaluation as a science, although at judgment of the panel and this would obviate the need for
the time it was not considered within that context. Formal statistics. The method is unique in that no direct judgement
descriptive analysis received its major impetus from “flavor was made concerning consumer acceptance of the product,
profile,” an approach that demonstrated it was possible to although most investigators assumed consumer acceptance
select and train individuals to describe the sensory proper- based on a result. An early recognized shortcoming of the
ties of a product in some agreed sequence, leading to action- method was its focus on the single sensory modality of fla-
able results, without requiring years as an apprentice to an vor. Subsequent methods would extend screening, training,
expert. The method attracted considerable interest and was a and evaluation to one or more additional modalities.
milestone in the development of the science of sensory
evaluation. It also was the source of much controversy, and
B. TEXTURE PROFILE
since then other descriptive methods have been developed.
Chronologically the next descriptive method of impor-
tance was the Texture Profile® method developed at the
A. FLAVOR PROFILE
General Foods Research Center (24, 25, 26). The method
The Flavor Profile® method described by Cairncross and was based on Flavor Profile, and extended descriptive analy-
Sjöstrom (21), Sjöstrom and Cairncross (22), and Caul sis into a second modality, texture. Brandt and co-workers
(23) was developed at the Arthur D. Little Co. and is the (24) defined a texture profile as “the sensory analysis of
only existing formal qualitative descriptive procedure. the texture complex of a food in terms of its mechanical,
The method utilizes a panel of four to six screened and geometrical, fat and moisture characteristics, the degree of
selected subjects who examine, discuss, and evaluate prod- each present and the order in which they appear from first
uct in an open session. Product attributes are scored for bite through complete mastication.” Several of the proce-
intensity in the order of their perception. The intensity dures introduced by Texture Profile raise behavioral issues
scale was: 0 ⫽ none,) (⫽ threshold,) (⫺1, 1 ⫽ slight, related to panel training.
1–2, 2 ⫽ moderate, 2–3, and 3 ⫽ strong. Once agreement If the panel is to be used also for flavor assessment, sub-
is reached on the description of the product, the panel ject screening includes tasks identical to those described for
leader summarizes and reports the results without appli- Flavor Profile. Texture screening tests were devised for hard-
cation of any statistical analysis. ness and viscosity ranking, and for geometric matching. As
Six subjects are selected for training based on technical many as 40 to 50 subjects may be screened, with 10 to 15
background and a series of screening tests, including basic selected for training. Full or partial denture wearers were
taste recognition and intensity ranking, odor identification not considered for panel membership.

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-13

Sensory Science: Methodology 57-13

Texture training included a two week orientation phase, and eliminates the opportunity to assess the importance of
a three month practice phase, and a three month expansion individual differences for understanding and generalizing
phase. A standard set of references was used to demon- results to consumers.
strate taste, aroma and mouthfeel sensations. The concept Interest in eliminating subject variability is, on the sur-
of the components of flavor profile was extended to include face, a reasonable (although not likely or practical) idea
what were described as mechanical, geometrical, and a cat- but it must not give way to procedures that sacrifice the
egory of “other” (moisture and fat) characteristics, inten- validity of the measurement system. The human is a living
sity, and the order of appearance for the texture properties. and changing organism, influenced by physiological and
An effort was made to describe the physical properties of psychological conditions that better support a concept of
texture and to provide a sensory definition for each des- expected variability. Products also are themselves variable.
cribed physical property. For these reasons, a panel is used rather than a single sub-
Initially, the texture terms and definitions were com- ject, and replication has become a mainstay in most cur-
piled, sorted, and categorized from both scientific/technical rent descriptive methods.
and popular terms, the latter from laboratory or field situ-
ations in which common usage terms were used to describe C. QUANTITATIVE DESCRIPTIVE ANALYSIS (QDA)
textural sensations. This led to the development of a clas-
Development of the Flavor and Texture Profile methods
Downloaded by [University of Newcastle] at 11:02 31 October 2015

sification of terms that were believed to encompass texture


sensations. Standardized rating scales were introduced for demonstrated a need for descriptive information and a for-
the mechanical and geometrical properties of texture per- malized procedure for screening and training subjects, and
ception, and these relied on commercially available products for evaluating products. They also stimulated interest and
to anchor each point of each scale. During training, subjects research on alternative descriptive methods that would cor-
were expected to learn these definitions and their associa- rect weaknesses previously identified: the extended amount
tions. The product evaluation procedure used a Flavor of time to select and train a panel, reliance on model systems
Profile approach with panelists discussing results in order (e.g., compounds in water) for screening subjects, an exper-
to reach a conclusion. imenter-derived descriptive language, the types of scales
Although the method was useful for focusing attention used, qualitative rather than quantitative information, the
beyond flavor and on the relative importance of texture, role of the panel leader, and the type and role of references.
it had numerous problems of its own. In developing the The Quantitative Descriptive Analysis (QDA) method, first
method, one objective was to eliminate subject variability described by Stone, Sidel, Oliver, Woolsey, and Singleton
through use of a standard vocabulary, and standard rating (27) and taught by the Tragon Corporation, addressed
scales and references as part of the training program. There behavioral, measurement, and quantitative weaknesses of
are inherent risks in experimenter-assigned attributes. While the profile methods, reflecting scientific and technical devel-
they may be meaningful to a chemist, one might question opments in psychology and the consumer products industry.
the perceptual meaning of these technical definitions to a The QDA method differs from the profile methods on
subject. Other subjects could ignore a perception because criteria of subjects recruited, screening procedure and
the characteristic was not among those selected for training. materials, subject training, testing, analysis, and reporting
Use of products as scale anchors presents their own set of for descriptive information.
problems. Products are not invariant; they change over time
as a function of marketing and other considerations. Another 1. Subjects
concern is the separation of texture from other sensory prop- Consumer users in the product category are recruited, and
erties of a product such as color, aroma, taste, and so forth. test results are then generalized to the consumer popula-
As a rule, perceptions are interdependent, and the exclusion tion. Experts and others having technical degrees in the
of a sensory modality from a scorecard does not eliminate product area are avoided as subjects because of the poten-
those perceptions. In effect, the subject is likely to use other tial for biased responses based on their technical knowl-
attributes to acknowledge these perceptions, and the visible edge of the product category, ingredients, and processes.
manifestation is increased variability and decreased sensi- About 25 subjects are recruited, and it is expected that 12 to
tivity. These other perceptions can influence the responses to 15 will successfully qualify and complete training.
the textural perceptions and vice versa.
Profile methods and related “expert” systems treat
2. Screening
subject differences as unwanted error, and these training
programs resort to behavior modification procedures in an Discrimination tests are used, and screening stimuli are
attempt to eliminate this source of variance. The panel selected from the product category rather than the simple
leader trains subjects to provide what the leader judges to stimuli in water solutions popular with the profile methods.
be the correct response to stimuli, and subjects are trained The closer the screening stimuli are to the test stimuli, the
to agree and repeat those responses in the presence of the more likely that successful performance will be obtained
stimulus. The approach ignores the reality of the situation, in the actual test. About 10 to 15 replicated pairs of product

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-14

57-14 Handbook of Food Science, Technology, and Engineering, Volume 2

stimuli are used, increasing in difficulty from the first pair and differences, and has become a standard in the food
(easy to discriminate) to the last pair (difficult to discrim- industry. A proprietary statistical package was made avail-
inate), covering all sensory modalities, and including any able for analyzing QDA data, and included in Tragon’s
differences known to be relevant to consumers. Correct training program for panel leaders. The statistical package
selection of the different product in more than about 65% uses one-way and two-way Analysis of Variance models
of the trials, and consistency in correct selection across in prescribed ways to assess subject performance. Current
replications are used as a guide for subject selection. innovations to the package include algorithms for assess-
ing crossover and magnitude interaction, scale use, and
3. Training multivariate analysis and mapping techniques. QDA data
is well suited to the recent advances in statistical analysis
QDA training focuses on language development and the and computer technology.
evaluation process, and provides subjects experience with
the range of products they will evaluate. Language devel-
opment is a consensus building process. The panel leader’s D. SENSORY SPECTRUM ANALYSIS
function is to keep the panel focused on developing objec- This method positions itself as a hybrid of the three methods
tive terms (in contrast to attitudinal terms) that can be described previously; however, it is primarily an extension
measured and obtain agreement amongst subjects on the
Downloaded by [University of Newcastle] at 11:02 31 October 2015

of the profile methods. Concepts first introduced for the


evaluation procedure. A typical QDA scoresheet includes Texture Profile method are extended back to the recruiting,
30 to 50 or more attributes covering all sensory modalities. screening, training, and testing for flavor panels as well.
The language development process contrasts with the pro- Separate training programs are required for flavor, texture,
file methods, which teach a preconceived technical language and each other modality. Training programs are quite
and use a prescribed set of “universal references” derived lengthy, a common element for all profile methods. Subjects
from the experimenter. QDA uses references selected are taught about sensory processes, similar to the orienta-
directly from the product category, and only on an as-needed tion programs found in the other profile methods. They
basis. Consequently, QDA derived attributes and consumer also are taught experimenter-assigned attribute vocabu-
perceptions take less time to obtain, with the entire train- lary and rating scales, the latter anchored at multiple points
ing process requiring between 5 to 10 hours, rather than with experimenter-assigned references. Subjects must learn
several months. the required response to selected stimuli, independent of
their individual perceptions, as an attempt to produce
4. Testing absolute scales. Teaching identical responses, independent
Subjects evaluate each product individually in a facility of individual perception, is possible; however, skeptics of the
equipped with test booths or partitions, or, for an extended method view this as a form of behavior modification and
use QDA, in their own homes following product usage. contrary to understanding and predicting consumer per-
Subjects typically evaluate 4 to 6 products, and as many ception. QDA concepts included in the Spectrum method
as 24 or more when part of product optimization research. include category-specific discrimination tests for subject
Balanced block serving designs are used and each subject screening, intensity scales (although more nominal or ordinal
provides 2 to 4 replicate evaluations. Subjects use a semi- than interval here because of the training method), exposure
structured line scale to score the perceived intensity for each to a broad range of product in the category, evaluation in
attribute for a product. The scale consists of a 6-inch individual test booths, and application of statistical analy-
(⬃15 cm) line, anchored ½ inch (⬃1.5 cm) from each end sis. The statistics are generic and less specific than those
to identify the direction (e.g., weak to strong) for each found in QDA and other recent developments in descrip-
attribute. This type of scale has no numbers and mini- tive methods. Curiously, variance data from panels trained
mizes word bias. It is very difficult for a subject to mem- in the Spectrum method has not adequately demonstrated
orize the location of previous judgements (assuring the zero result necessary to support a claim of absolute
greater independence of judgements from replication to scales or counter the concept of individual differences.
replication and product to product), and preserves the
interval character of the scale for data analysis. E. FREE-CHOICE PROFILING
Williams and Langron (28) described a radically different
5. Data Analysis
descriptive method that required no subject screening and
QDA was the first trained panel method to insist on, and training, where each subject could use their own words
provide for, statistical analysis of subject performance and and scales to describe products. Generalized Procrustes
product differences. Line scale responses are converted to analysis, an iterative type of factor analysis, is used to
numerical values from 0 to 60 prior to analysis. A unique force fit individual data into a common space. While this
method of graphing the data, QDA “spider graphs,” was type of summary analysis can show differences between
developed that readily communicated product similarities very different products, it misses smaller, and possibly

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-15

Sensory Science: Methodology 57-15

important, differences. The method does not demonstrate Time-intensity studies have been reported for the taste,
any substantially unique advantages and may be better aroma, and texture of foods, and for a variety of products
suited to classification and comparison to other consumer including bread, chewing gum, beer, wine, ice cream, chili
procedures than to trained panel procedures. pepper, gelatin, and chocolate. Commercial interest in syn-
thetic sweeteners in the 1970s prompted significant pub-
F. OTHER METHODS lished and unpublished research into the temporal effects
of these sweeteners compared to sucrose.
Descriptive information is recognized as important, if not
crucial, as a means for providing a comprehensive product
assessment. Descriptive panels are used in several venues A. MEASUREMENT
including, and not limited to, product development, qual- Interest in time and intensity relationships in sensory eval-
ity control, advertising claim support, and food science uation is evident by the almost universal practice of meas-
education. Different needs have given rise to descriptive uring aftertaste or aftereffect in descriptive analysis and
panel variations other than the methods described above. consumer research. Subjects in time-intensity studies typi-
Many descriptive methods are generic and retain in varying cally receive instruction and practice with the test procedure,
degree elements of the core methods; a few other methods scale, and any special apparatus used for data collection,
have established, or attempted to establish, themselves as
Downloaded by [University of Newcastle] at 11:02 31 October 2015

and may be screened for their ability to detect differences


different. Larson-Powers and Pangborn (29) introduced the for the attribute or attributes measured. Under these condi-
use of a reference combined with a difference from refer- tions, the subjects are viewed as a special type of descriptive
ence scale. The method required simultaneously available analysis panel.
stimuli, reducing the number of different products that The simplest measurement practice involves scoring
could be evaluated in a session. The developers of the Flavor the perceived intensity for overall aftertaste or aftereffect
Profile introduced a new version called Profile Attribute (i.e., sensation remaining in the mouth), or for a specified
Analysis (30) that included 7-point intensity scales allowing attribute, as a last task during routine product evaluation by
for statistical treatment of the responses. Stampanoni (31) an ongoing descriptive panel. The panel is instructed to first
introduced “Quantitative Flavor Profiling” as a hybrid wait a prescribed amount of time (30 to 60 seconds or more
method based on Flavor Profile and QDA. depending on the product) after completing other attribute
evaluations. This procedure can be expanded to include
intensity measures for several specific attributes obtained
VII. TIME-INTENSITY SCALING
at one or more designated points in time during the evalu-
The temporal aspects of the sensory experience during food ation. This approach is appealing because it uses an ongo-
and beverage consumption are apparent to even the casual ing panel, can measure several attributes at selected points in
observer. Not unexpectedly, perceptual changes occur as time, and special data collection equipment is not required.
the physical properties of a food change during chewing, or There is a practical limit to the number of attributes that
while in the mouth (e.g., ice cream and butter). Perceptual can be measured during a time interval, as it takes time to
change also can occur when the chemical properties of a score attributes. This and related issues should be resolved
stimulus remain constant, as with taste, odor, and certain during training. The primary disadvantage of this approach
mouth sensations (e.g., burning sensation from chemical hot it that measurement occurs at discrete points in time,
stimuli). Early psychophysical research into taste adapta- whereas, the perceived change may be continuous. Whether
tion included time as a variable, measuring changes in taste the perceived change is different for different subjects is
threshold over time for simple adapting solutions. Hahn’s more a data analysis issue than one of data collection.
(32) results demonstrated threshold differences based on the Continuous measurement of the change in perceived
concentration of the adapting solution and for the basic intensity requires an apparatus that simultaneously tracks
tastes measured (salt, sugar, bitter, and acid). The research time and the subject’s intensity scores. While providing the
also concluded that subjects complete recovery to the adapt- advantage of continuous measurement, the apparatus usu-
ing solutions was not reached within the 30 seconds allowed, ally limits the number of attributes that can be measured
and that there were individual differences. Individual dif- to one or two, and based on the apparatus requires varying
ferences in time-intensity research continue to be reported; degrees of physical involvement by the subject. Pen type
for example, van Buuren (33), using principal components strip-chart recorders were among the earliest devices used
to analyze time-intensity (TI) curves, concluded “TI curves for continuous measurement in time-intensity studies (34).
appear to be determined by characteristics that are related to The recorder moves chart paper at a constant rate and an
the judge and that may be unrelated to the product.” Given intensity scale is fixed on the recorder perpendicular to the
that most time-intensity studies involve small numbers of direction of chart flow. The subject continuously records
subjects and data are averaged, individual differences are perceived intensity for the selected attribute by keeping
a vexing concern for generalizing time-intensity results to a the pen in contact with the chart paper, moving the pen in
larger consumer population. the direction of increasing or decreasing intensity until the

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-16

57-16 Handbook of Food Science, Technology, and Engineering, Volume 2

perceived intensity for the attribute has returned to base- that make the use of statistics especially challenging. First,
line. Data from early generation strip chart recorders was there are many different statistical procedures from which to
handled much like that from similar tests using paper bal- choose, and new tests are regularly introduced and described
lots. The subject’s intensity score at several points in time in the literature. Second, statisticians as well as sensory sci-
were hand measured and converted to numerical value for entists differ in their view about how responses should be
data analysis. This was a labor-intensive process, depend- treated, which type of analysis is more or less appropriate
ing on the number of time points examined. and useful, the extent of any data transformation, and so
Desktop computers linked to a joystick (35), or simi- forth. Third, software packages and the power of the PC
lar device, for measuring perceived intensity introduced make all methods for analysis easily obtainable in terms
significant improvements in the time required to collect, of time and cost, but that does not make them appropriate.
enter, and analyze data, and simplified the subject’s task This section provides an introduction to the use of sta-
compared to strip chart recorders. These advances and tistics as a guideline for making sensory and business deci-
developments in statistical analysis and graphing software sions about products of interest. The focus is on statistical
that could be used with time-intensity data encouraged applications rather than statistical theory and the proofs,
greater use of the method for sensory research. or various algorithms that would take far more than the
space allocated here for this topic. The interested reader
Downloaded by [University of Newcastle] at 11:02 31 October 2015

B. ANALYSIS can find that information and worked examples in books


covering the general topic of sensory evaluation and other
Time-intensity studies typically involve one or two attrib- books that focus on statistics for sensory evaluation (17,
utes, and the results are readily graphed on x-y coordinates, 36, 37). Sensory decisions based solely on the statistical
with attribute intensity on the y-axis and time on the x-axis. conclusions are the exception, not the rule. The statistics
Curves may represent individual data, averaged data for that we describe will focus primarily on those analyses
the panel, or group data for subjects having similar results. found to be useful for the behavioral sciences; e.g., vari-
Measures of central tendency, such as means or medians, ance measures, and will focus less on statistics more typ-
are used to construct data points for selected time periods, or ically used in the engineering and physical sciences.
data may be otherwise transformed to reduce or eliminate
subject differences. Data points are then connected to cre-
A. DESCRIPTIVE STATISTICS
ate a time-intensity (TI) curve. Selected curve parameters
or discrete points are compared directly or analyzed for Sensory responses may take the form of discrete or scaled
statistical significance. Maximum intensity (Imax), time to data. An example of discrete data is found in discrimination
maximum intensity (Tmax), and duration (return time to results where subjects select a product as similar or differ-
baseline intensity) are popular discrete points used for ent compared to one or more other products. An example
comparing TI results for different products, and suitable of scaled data is found where subjects score individual sen-
analysis of variance and multiple range tests are then per- sory attributes. In this case scoring may be a ranking, or
formed to determine significance for the observed differ- rating on one form or another of an interval or ratio scale.
ences. An examination of the literature on this topic reveals Descriptive data provides a summary of the individual
views espousing and challenging a variety of curve param- scores, and as such, is more manageable than the original
eters worthy of consideration, such as area under the curve, raw data. However, summary statistics by their very nature
peaks, plateaus, slopes, and so forth. Several investigators can also mask or distort information and the sensory scien-
have proposed combinations of these and other measures, tist must be aware of this potential problem. Descriptive sta-
and developments are still forthcoming in this area. The tistics are grouped into measures for central tendency and
interested reader is directed to Lawless and Heymann (1) for dispersion, and are often displayed in tables and graphs.
additional detail about TI research in sensory evaluation
and the different analysis methods proposed for the data. 1. Central Tendency
One of the biggest challenges facing the experimenter
is how to deal with a subject’s error of anticipation; i.e., The mean, mode, and median are important measures for
providing responses based on the subject’s guessing the summarizing individual data.
purpose of the test. a. Mean
The arithmetic mean for scaled data is the sum of all scores
divided by the number of scores that have been summed.
VIII. DATA ANALYSIS
The arithmetic mean is the most widely used measure in
Statistics is an essential part of the sensory evaluation sensory evaluation. It can be distorted by extreme scores,
process, providing a means of summarizing information to in which case other means such as the harmonic mean or
reach conclusions about subject performance and product geometric mean may be better choice. Rank means are pos-
difference. For sensory evaluation there are several issues sible when rank order scales are used.

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-17

Sensory Science: Methodology 57-17

b. Mode B. INFERENTIAL STATISTICS


The mode is the response category that contains the largest
This category of statistics is used by the sensory
number of responses. This measure is not influenced by
researcher to determine the risk associated with declaring
extreme scores as is the arithmetic mean.
that an observed difference in a test represents a real dif-
c. Median ference that can be generalized to other test populations, or
The median is the category containing the 50th percentile, to a larger consumer population. Using appropriate statistics
dividing an ordered distribution in half. This measure is does not eliminate the risk of making a wrong decision;
also not influenced by extreme scores. nonetheless such errors can be minimized through proper
experimentation and analysis. There is not sufficient space
in this chapter to include discussion about probabilities
2. Dispersion
and hypothesis testing; therefore, the interested reader is
The range, variance and standard deviation are the three directed to the aforementioned publications (17, 33, 34).
most frequently used dispersion statistics reported in sen- We will limit our discussion to those inference statistics
sory tests. used most frequently by sensory practitioners.

a. Range
Downloaded by [University of Newcastle] at 11:02 31 October 2015

The range refers to absolute difference between the lowest 1. Non-Parametric


and highest scores of an ordered distribution. The range is
This category of statistical tests is used when data from
easily influenced by one or more extreme scores, and by
groups or products to be compared are in the form of counts
itself has limited value. It is more informative when com-
(i.e., frequencies), percents, or ranks. These are distribu-
bined with measures of central tendency, variance, and a
tion free statistics, with no constraint regarding the shape
frequency distribution.
(e.g., symmetrical bell shape Gaussian curve referred to as
b. Variance the normal curve) of the distribution. The number of non-
The variance provides a measure of deviation or dispersion parametric tests is too large to discuss here, and the inter-
from a mean. To calculate the variance for a data set ested reader will find greater discussion and useful
requires that each score is subtracted from the mean, each references in the sensory statistics literature cited above.
difference from that operation squared (to eliminate nega- We mention only a few of these tests below.
tive numbers), these squared differences summed, and then
a. Binomial tests
divided by the number of scores minus 1. The calculated
Data from forced choice tests such as those described for
variance resulting from this process is displayed as S 2 to
most discrimination tests can be analyzed for statistical
represent sample data rather than population data.
significance using tables based on the binomial expan-
c. Standard deviation sion. This has been described previously in this chapter.
The standard deviation is the square root of the variance
b. Chi-square (χ2)
and displayed as S to represent sample data. As may be
This is a useful statistic for determining whether the
expected, the standard deviation will always be a smaller
observed number of responses is significantly different
value than the variance, further reducing the impact of
from an expected number (derived from the total number
extreme scores.
of responses). There are several variations of the formula
that can be used based on how the data are collected and
3. Frequency Distributions the number of classifications being compared. In sensory
evaluation the chi-square test is often used to determine
Frequency is the number or percent response for each whether two response distributions are significantly dif-
response category. Frequency data are typically displayed ferent, and as a preliminary test for isolating and compar-
as tables or graphs, which provide useful information about ing two response categories from among several. It is used
how well other measures, such as the mean and range, rep- more in sensory consumer tests and less so in sensory dis-
resent the data. For example, a flat distribution would indi- crimination and descriptive tests.
cate that the mean value is not a good summary description
for the data. Sensory specialists routinely examine fre- c. Other non-parametric tests
quency distributions for each product and serving order The aforementioned texts include useful descriptions for
combination, and different groups of subjects to identify several other non-parametric tests worthy of note for sen-
important trends in the data. Products having similar sory analytical tests. Those statistical tests include, and
shape distributions may readily be included in a variety of are not limited to; the McNemar test, Cochran Q test,
statistical significance tests, whereas those with very dif- Wilcoxon test, Mann-Whitney U test, Friedman test, and
ferent shape distributions may not. the Kruskal-Wallis test.

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-18

57-18 Handbook of Food Science, Technology, and Engineering, Volume 2

2. Parametric of the calculated F are published in most statistics books;


most statistical packages for sensory and behavioral data
This category of statistical tests is reserved for data that
include the AOV analysis and exact probabilities for the
satisfies the constraints of the normal curve, which includes
calculated F.
equality of the scaled intervals. All things being equal,
which they seldom are, parametric tests are viewed as more c. Multiple range tests
powerful than non-parametric tests. When response data Multiple range tests are used following an AOV to deter-
are clearly nominal (e.g., counting and classification data) or mine from among a set of means for a significant variable
ordinal (e.g., ranks), non-parametric tests are recommended. which ones are significantly different. Different range tests
Parametric tests are appropriate when data are clearly inter- reflect what is to be compared and the respective authors’
val or ratio. Having said this, the reader is cautioned that preference for controlling the various error rates possible
most of the scales used in sensory evaluation have sufficient when making multiple comparisons for a single data set.
interval characteristics to permit judicial use of paramet- Too conservative a test will make it difficult to find statisti-
ric statistics. Underlying statistical assumptions and theory cal significance and result in more Type II errors, whereas
is just that, assumptions and theory. In practice, results may a less conservative test may result in false differences (i.e.,
be reliable, valid, and useful even if they do not perfectly Type I error). The sensory scientist is cautioned here that the
satisfy underlying assumptions. In this event, it may be more underlying theoretical assumptions for a multiple range
Downloaded by [University of Newcastle] at 11:02 31 October 2015

beneficial to modify the theory and assumptions, rather than test may not agree with practical experience using that test.
forgo the application. Where this is the case, paying attention to the risk one can
still proceed cautiously. Frequently used multiple range
a. t-test tests include; Fisher’s LSD (least significance difference),
The t-test is used to determine whether the means for two Dunnett’s test, Duncan’s test, some tests by Tukey, the
products are significantly different. Different formulae are S-N-K (Student-Newman-Keuls) test, the Scheffé test, and
available to accommodate various conditions such as: inde- the Bonferroni test. Whichever test is selected, the results
pendent and dependent observations, large and small N, must be examined to determine whether they agree with the
proportions, and equal and unequal number of observations. knowledge and expectation the researcher has about the
The t-test is not the best choice for sensory tests where the products and the sensory test.
same subjects evaluate both products, the analysis-
of-variance can be substituted and it extracts more from the
error variance. The t-test also is inappropriate where more
C. CORRELATION AND REGRESSION
than two products are to be evaluated and statistically com- Sensory scientists are often interested in the relationships
pared to one another. This constitutes multiple t-tests of the between different sets of data. The interest may be bet-
same data, a common error. ween different groups of subjects, products, attributes, or
combinations thereof. Because correlation does not imply
b. Analysis of variance (AOV) causation, summary statistics and graphs depicting asso-
The analysis of variance is the primary statistical test used ciation are useful for data reduction, substitution, predic-
for scaled sensory data. It is a versatile statistical procedure tion, and generally improving understanding about
that readily accommodates multiple products, treatment variables. As with inference testing, different statistics are
conditions and groups of subjects. It is used to determine used for different levels of measurement (e.g., nominal,
whether an effect is statistically significant, and when a rank order, or continuous).
difference noted is followed by a multiple range test to These measures are used to examine and better under-
determine which means are different. Several AOV models stand sensory and instrumental relationships, and in opti-
are extensively used in sensory evaluation; these include the mization research where sensory attributes are used as
one-way for independent means, the two-way for depend- predictor variables for consumer acceptance.
ent means, treatment x levels for different treatment con-
ditions (e.g., x-levels of different variables), and split-plot
1. Correlation
models for different subject groups evaluating the same
and different treatments. To understand a simple two-variable correlation it is best
The primary challenges for selecting the best AOV to first graph results as a scatter plot with the data from
model are to select one that is appropriate for the specific one variable on the X-axis and data from the second variable
test conditions, to accurately account for all the sources of on the Y-axis. For example, descriptive scores for sweet-
variance in the test, and to select the appropriate error term ness and bitterness for a set of products can be graphed
for testing the significance of an effect. Statistical signifi- and a correlation for the relationship between the two attrib-
cance in the AOV is based on a calculated F ratio consisting utes calculated. The Pearson product moment correlation
of an effect variance in the numerator and an error variance coefficient is used for continuous scaled data and is depicted
in the denominator. Probability tables for the significance by r. Calculated r-values range from ⫺1 to 0 to ⫹1. The

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-19

Sensory Science: Methodology 57-19

closer to 0, the less is the association between the two vari- 2. Discriminate Analysis
ables. The minus sign depicts an opposite relationship,
Discriminate analysis is a methodology for finding linear
whereas a plus sign depicts a directionally similar rela-
combinations of the independent variables that can act as
tionship; unity represents a perfect correlation. Pearson’s is
scoring functions to estimate to which of several classifi-
useful only for linear data, curvilinear data can produce low
cation categories an observation belongs. Sensory scientists
values approaching 0, misleading the casual observer to
use this and related procedures (e.g., canonical correlation)
conclude the variables are unrelated. Correlation can be
in optimization research to identify lifestyle, attitude, and
determined for non-continuous scaled data as well; an
classification information that best identify membership
example is Spearman’s rank order correlation coefficient
in different consumer preference groups.
(ρ Greek lowercase rho).
3. Principle Components Analysis
2. Multiple Correlation (R)
This data reduction technique transforms the original set
Multiple Correlation is used to assess the degree of asso-
of variables into as few linear combinations as possible to
ciation between a dependent variable and a set of predictor
explain as much of the total variation in the original data.
variables. In product optimization research several sensory
The linear combinations, identified as the factors or compo-
attributes (independent variables) may be included as
Downloaded by [University of Newcastle] at 11:02 31 October 2015

nents, are independent from all other factors. The method


important predictors of consumer acceptance (dependent
is often applied to sensory descriptive panel data to reduce
variable), and the multiple correlation (R) describes the
the number of attributes, or to identify a smaller set of
degree of association between the dependent and independ-
independent attributes to include as independent variables
ent variables. Several books have been published on the
in developing multiple regression models for optimization
topic of Multiple Regression/Correlation (MRC), and the
research. It is also used in optimization research to identify
interested reader is directed to Cohen and Cohen (38) for an
products that form independent groups based on sensory
in-depth discussion.
attributes.
3. Regression
4. Factor Analysis
Regression is a general term applied to equations that fit a
Similar to principle components analysis, factor analysis
line to observed data points. Simple linear, non-linear, and
is a data reduction technique, except it focuses on that part
multiple regression are possible, and the resulting regres-
of the total variation that a variable shares with other vari-
sion line is often used to predict values of the dependent
ables in the set. Some sensory researchers prefer to build
variable (y) from values of the independent variable (x). The
optimization models using factor scores rather than attrib-
sensory scientist will use regression equations, for exam-
utes identified through principle components analysis.
ple, to predict perceived intensity for an attribute based on
However, such a result has limited value when applying the
ingredient concentration. In optimization research, multi-
information to make product change, for example.
ple regression is used to predict consumer liking from a
combination of sensory attributes and their intensities.
5. Cluster Analysis
Analytical measures may also be used as predictor vari-
ables for acceptance in optimization, alone or in conjunc- This data reduction technique is used to find a smaller num-
tion with sensory measures. ber of groups whose members have elements more similar
to one another than to members of another group. Cluster
D. ADDITIONAL MULTIVARIATE METHODS analysis is used in sensory optimization research to identify
unique consumer preference groups, and it is used in the
Several tests in addition to those already described are used
same way in descriptive analysis to identify unique subject
in sensory evaluation to simultaneously examine multiple
groups. In the latter situation, uniqueness may indicate the
variables. Multivariate analysis often provides understand-
need for additional panel training.
ing of important relationships imbedded within the data
and not readily observable by other means. The interested
reader will find that the book by Dillon and Goldstein (39) IX. VISUAL PRESENTATION
offers a useful integration of theory, explanation and illus-
Visual presentations are an important part of understanding
tration for the topic.
and communicating data. They reveal events and trends in
data not readily observed any other way, and they simplify
1. Multivariate Analysis-of-Variance (MANOVA)
results for less technical and non-statistical audiences (40).
This analysis is used to test the significance of a set of vari- Increased availability of desktop computers capable of man-
ables. A good example would be descriptive panel data aging large amounts of data has proven to be of great bene-
where several attributes are evaluated for a product set. fit to sensory scientists. The sensory scientist can stretch,

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-20

57-20 Handbook of Food Science, Technology, and Engineering, Volume 2

shrink, rotate, organize and reorganize large and small Smooth Mf


amounts of data with a simple keystroke or two, and in
doing so examine countless relationships among subjects,
attributes and products. In addition to the large number of
generally available analysis and graphing software pro-
Warm Mf Body Mf
grams, such as SAS and SAS/JMP (SAS Institute Inc., Cary,
NC), S-Plus (MathSoft Inc., Seattle, WA), SPSS and BMDP
(SPSS Inc., Chicago, IL), custom programs have been
developed by several companies for application in sensory
evaluation. For the latter, the interested reader is directed to:
Camo, Inc. Corvallis, OR; Compusense, Guelph, Ontario,
Canada; Sensory Computer Systems, Lancaster, PA; and
Tragon Corporation, Redwood City, CA. A few examples of
graphs useful to the sensory scientist are described below.

Puckery Mf Biting Mf
A. HISTOGRAMS
Downloaded by [University of Newcastle] at 11:02 31 October 2015

Frequency histograms, together with central tendency and


variance measures, provide important information about
sensory responses to products and how well the data satisfy Chalky
assumptions required for other statistical tests, and allow Wine D coating
for comparisons between products. Figure 57.1 illustrates Wine G
a different Product by Serving Order distribution for Wine K
24 subjects, where each subject evaluated both products
FIGURE 57.2 Typical “spider” plot for selected mouthfeel
(i.e., 12 subjects per serving sequence). The means are lower attributes. Intensity scores are measured from the center out to
for both products in the second serving order, demonstrat- the point where the product crosses. The attributes are spaced
ing a possible order bias, a not uncommon finding in sen- equally around the center point. Bars on spokes represent the
sory studies, and the primary reason for using balanced magnitude of difference for that attribute to be significant using
serving orders. The bimodal distribution and variance the LSD test.

Product A- Product A- Product B- Product B-


1st position 2nd position 1st position 2nd position
6
Mean = 7.92 Mean = 5.08 Mean = 6.17 Mean = 5.25
Variance = 1.17 Variance = 8.26 Variance = 1.97 Variance = 1.48
5

4
Frequency

0
1 5 10 1 5 10 1 5 10 1 5 10
Sweetness intensity
(N = 12; Scale = 1 to 10)

FIGURE 57.1 Illustration of distribution histograms for examining serving order effect. The graph represents 24 subjects that
evaluated two products in two serving orders. Twelve subjects evaluated Product A first, and 12 subjects evaluated Product B first.

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-21

Sensory Science: Methodology 57-21

differences for Product A when served in the second order descriptive panel data. The maps often group well liked
would alert the sensory researcher that it may be risky, or products together with sensory attributes and consumer
even inappropriate, to conduct statistical tests for this information (e.g., lifestyle and attitude) identified as
result if homogeneity of variances is an issue. highly related to those product preferences. In the case of
descriptive panel data, it is often helpful to visually dis-
B. QDA SPIDER CHARTS play products by attribute results. Figure 57.3 is such a
graph, and demonstrates how products may differ, or
Figure 57.2 represents a popular way for graphing
group, based on several attributes across different modal-
descriptive panel results (2,27). Figure 57.2 displays results
ities. Here we see that Wines K and D are scored high for
only for mouthfeel from a study that included additional
several different attributes; whereas, Wine G falls between
modalities. Examining Figure 57.2, we readily see product
the two. We would conclude that Wine K is a strong and
differences for individual attributes, relationships among
high impact product in several areas.
the attributes, and the overall tendency for a product to score
Figure 57.4 is an example of a density map based on a
low (Wine D), high (Wine K), or in the middle (Wine G)
cluster analysis of consumer acceptance scores for a broad
compared to other products evaluated.
range of products of the type typically included in a cate-
gory review optimization. The cluster map reveals three
C. SENSORY MAPS
Downloaded by [University of Newcastle] at 11:02 31 October 2015

reasonably unique preference groups embedded within


Mapping is frequently used for displaying results of opti- the aggregate population. An appropriate follow-up sen-
mization research involving consumer acceptance and sory strategy is to use multiple regression techniques

0.6

0.4

Metallic Ar
Mouthdry Af t
Sour Aft
Coating Aft
Discriminant function 2 (21.2%)

Green Fl Duration Aft Body Mf


D Grape Fl Fruit Aft K
0.2 Overall Af t
Earthy Ar
Smooth Mf Medicinal Ar Chalky coating
Brown tint
Oak Ar
Fruity Fl
Peppery Fl
0.0
Grape Ar Smoky Fl
Perfumey Ar
Vanilla Fl Bitter Fl Biting Mf

–0.2 Tears
Clear Ap

–0.4
G

–0.4 –0.2 0.0 0.2 0.4 0.6


Discriminant function 1 (78.8%)

FIGURE 57.3 Attribute factor loading for trained panel evaluation of three wines; D, G, and K.

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-22

57-22 Handbook of Food Science, Technology, and Engineering, Volume 2

Cluster density map

Group 2

Group 3

Group 1
Downloaded by [University of Newcastle] at 11:02 31 October 2015

Preference group 1

Preference group 2
Group centroids
Preference group 3

FIGURE 57.4 Cluster Analysis density map for aggregate acceptance scores. Three unique preference groups were identified.

2
Wine D Wine D Wine K
Wine K
Wine D
Wine D Wine D Wine K
Wine K Wine K
Discriminat axis 2 (21.2%)

Wine D
Wine K
Wine K
Wine D Wine G
0 Wine K
Wine G Wine D Wine K

Wine G
Wine D Wine G
Wine G
Wine D

Wine G Wine K
Wine G G
Wine
–2
Wine G
Wine G

–4

–2 0 2 4
Discriminate axis 1 (78.8%):

FIGURE 57.5 Discriminate plot for three wines across sensory attributes scored by each of 10 subjects. Arrows show two subjects that
differ compared to the other subjects for their evaluation of Products G and K. Filled arrow is Subject #167; open arrow is Subject #542.

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-23

Sensory Science: Methodology 57-23

Sensory Map

Grape flavor
Fruity aftertaste
Consumer liking

Current control
Market leader

Body Grape aroma Clear appearance


Coating
Factor 2

Prototype A
Perfumy aroma
Competitor
Smoky flavor
Downloaded by [University of Newcastle] at 11:02 31 October 2015

Tears
Bitter flavor
Prototype B

Biting mouthfeel

Factor 1

FIGURE 57.6 Sensory map integrating consumer liking and sensory panel attributes for selected products.

TABLE 57.1 obtainable and provide important insights about products


Factor Analysis Summary for Consumer Liking and and subjects.
Sensory Attributes Figure 57.6 is a sensory map that includes consumer
Factor 1 Factor 2 Factor 3 acceptance and sensory panel results. The spatial relation-
Clear Appearance 0.935
ships displayed in these types of maps (41) provide use-
Grape Aroma 0.510 ful information to product developers and marketing
Coating ⫺0.804 ⫺0.554 researchers. In this case, the map identified which sensory
Body ⫺0.909 attributes were most associated with which products, and
Grape Flavor 0.732 0.557 that there is a development opportunity to optimize con-
Fruity Aftertaste 0.702 sumer liking for the product category. Sensory maps are
Consumer Liking 0.533 typically derived from Factor Analysis (FA) or Principal
Tears ⫺0.551 Component Analysis (PCA) output, similar to that seen in
Bitter Flavor ⫺0.557 Table 57.1.
Biting Mouthfeel ⫺0.888
Perfumy Aroma 0.950
Smoky Flavor 0.873 X. CONCLUSION
Total
% Variance Explained: 41.13 21.92 19.83 82.88 Sensory evaluation is a valuable resource for the food sci-
entist, and has yet to realize its full potential within the
food industry. This is partly due to a lack of awareness of
to develop sensory models that best fit each unique pref- what it can do and how to best organize and operate sen-
erence group as well as one or more models for the aggre- sory capabilities. The science is dynamic and its applica-
gate population. tions are almost endless, especially as it relates to its role
Figure 57.5 is based on a discriminate analysis of indi- in consumer research. The sensory sciences outlook is
vidual subject scores for the three wines. This information bright as more academic institutions teach and research
provides one way to visually identify how well subjects the topic, and more qualified professionals enter the field.
discriminated among the products. Examining Figure 57.5, In recent years there has been a significant increase in the
we see that two subjects differed most among the panel use of sensory methods as multinational and domestic
regarding scores for Wines G and K. This information can companies introduce these procedures on a global basis.
be used to determine when additional training for subjects This trend will continue, creating new opportunities and
is required, or when a subject may be removed from a panel. demands for the science. Future success will depend in
In either case, these types of visual displays are readily large part on how well sensory professionals and those

© 2005 by Taylor & Francis Group, LLC


DK9848_C057.qxd 04/Nov/05 04:46 PM Page 57-24

57-24 Handbook of Food Science, Technology, and Engineering, Volume 2

that teach and employ them understand that sensory eval- of Foods. 2nd ed. (JR Piggott, ed.). London: Elsevier
uation is about perception; and how best to measure, ana- Applied Science, 1988.
lyze and communicate sensory information. 21. WE Cairncross, LB Sjöström. Flavor profile — a
new approach to flavor problems. Food Technol 4:
308–311, 1950.
22. LB Sjöström, SE Cairncross. The descriptive analysis of
REFERENCES flavor. In: DR Peryam, FJ Pilgrim, MS Peterson. eds.
Food Acceptance Testing Methodology. Washington, DC:
1. HT Lawless, H Heymann. Sensory Evaluation of Food: Natl Acad Sci — Natl Res Counc 1954 pp 25–30.
Principles and Practices Gaithersburg, Maryland: Aspen 23. JF Caul. The profile method of flavor analysis. Adv Food
Publishers, Inc., 1998. Res 7:1–40, 1957.
2. H Stone, JL Sidel. Sensory Evaluation Practices, 3rd ed. 24. MA Brandt, E Skinner, J Coleman. Texture profile
San Diego: Academic Press, 2004. method. J Food Sci 28:404–410, 1963.
3. MA Amerine, RM Pangborn, EB Roessler. Principles of 25. AS Szczesniak. Classification of textural characteris-
Sensory Evaluation of Food. New York: Academic tics. J Food Sci 28:385–389, 1963.
Press, 1965. 26. AS Szczesniak, MA Brandt, HH Friedman. Development
4. DR Peryam, FJ Pilgrim, MS Peterson, (eds.). Food of standard rating scales for mechanical parameters of
Acceptance Testing Methodology. Washington D.C.: texture and correlation between the objective and the sen-
Downloaded by [University of Newcastle] at 11:02 31 October 2015

Natl. Acad. Sci — Natl. Res. Counc, (1954). sory methods of texture evaluation. J Food Sci 28:
5. Anonymous. Minutes of Division Business Meeting. 397–403, 1963.
Institute of Food Technologists — Sensory Evaluation 27. H Stone, J Sidel, S Oliver, A Woolsey, RC Singleton.
Division, IFT, Chicago, Illinois, 1975. Sensory evaluation by quantitative descriptive analysis.
6. MC Bourne. Food Texture and Viscosity: Concept and Food Technol 28:24–34, 1974.
Measurement. New York: Academic Press, 1982. 28. AA Williams, SP Langron. The use of free-choice pro-
7. A Noble. Instrumental analysis of the sensory properties filing for the evaluation of commercial ports. J Sci Food
of food. Food Technol 29:56–60, 1975. and Agric 35:558–568, 1984.
8. HG Schutz, VG Vely, RB Iden. Relation of odor to 29. N Larson-Powers, RM Pangborn. Descriptive analysis
physicochemical properties of compounds (Interim. of the sensory properties of beverages and gelatins con-
Reprot). QM Food and Container Institute for the Armed taining sucrose or synthetic sweeteners. J Food Sci 43:
Forces. (Contract No. DA-19-129-QM-1500), 1961. 47–51, 1978.
9. WF Wilkens, JD Hartmann. An electronic analog for the 30. JE Hanson, DA Kendall, NF Smith. The missing link:
olfactory process. J of Food Sci 29:372–378, 1964. correlation of consumer and professional sensory
10. PN Bartlett, JM Elliott, JWGardner. Electronic noses descriptions. Beverage World 102:108–116, 1983.
and their application in the food industry. Food Technol 31. CR Stampanoni. The quantitative flavor profiling tech-
51: 44–48, 1997. nique. Perfumer and Flavorist 18:19–24, 1993.
11. RN Bleibaum, H Stone, T Tan, S Labreche, E Saint- 32. H Hahn. Die adaptation des geschmackssinnes.
Martin, S Isz. Comparison of sensory and consumer Z Sinnesphys 65:105–145, 1934.
results with electronic nose and tongue sensors for apple 33. S van Buuren. Analyzing time-intensity responses in sen-
juices. Food Quality and Preference 13:409–422, 2002. sory evaluation. Food Technol 46:101–104, 1992.
12. NS Lewis. The Caltech Electronic Nose Project. 34. N Larson-Powers, RM Pangborn. Paired comparison
Engineering & Science, No 3, 1996. and time-intensity measurements of the sensory proper-
13. G. Taubes. The Electronic Nose. Discover 40–50, 1996. ties of beverages and gelatins containing sucrose or syn-
14. L Malcolmson, L Jeffrey, DD Sharma, PKW Ng. thetic sweeteners. J Food Sci 43:41–46, 1978.
Relationship between sensory clarity and turbidity 35. JX Guinard, RM Pangborn, CF Shoemaker. Computer-
values of apple juice. Canadian Institute of Science and ized procedure for time-intensity sensory measurements.
Technology Journal 22:129–132, 1989. J Food Sci 50:543–546, 1985.
15. J Eggert, K Zook, eds. Physical Requirement Guidelines 36. MC Gacula Jr, J Singh. Statistical Methods in Food and
for Sensory Evaluation Laboratories. ASTM STP 913. Consumer Research. Orlando: Academic Press, 1987.
Am Soc Test Mater, Philadelphia, Pennsylvania, 1986. 37. MC Gacula Jr. Design and Analysis of Sensory Optimiza-
16. EB Roessler, RM Pangborn, JL Sidel, H Stone. tion. Connecticut: Food and Nutrition Press, 1993.
Expanded statistical tables for estimating significance in 38. J Cohen, P Cohen. Applied Multiple Regression/Correla-
paired-preference, paired-difference, duo-trio and trian- tion Analysis for the Behavioral Sciences. 2nd ed. New
gle tests. J Food Sci 43:940–943, 1978. Jersey: Lawrence Erlbaum Associates, 1983.
17. M O’Mahony. Sensory Evaluation of Food: Statistical 39. WR Dillon, M Goldstein. Multivariate Analysis: Methods
Methods and Procedures. New York: Dekker, 1986. and Applications. New York: John Wiley & Sons, 1984.
18. DM Ennis. The power of sensory discrimination meth- 40. WS Cleveland. Visualizing Data, Summit, New Jersey:
ods. J. Sensory Studies 8:353–370, 1993. Hobart Press, 1994.
19. DM Ennis, K Mullen. Theoretical aspects of sensory 41. K Greenhoff, HJH MacFie. Preference mapping in prac-
discrimination. Chem. Senses 11:513–522, 1986. tice. In: Measurement of Food Preferences. (HJH
20. JER Fritjers. Sensory difference testing and the meas- MacFie and DMH Thompson, eds.). Blackie, Academic
urement of sensory discriminability. In Sensory Analysis and Professional. 137–166, 1994.

© 2005 by Taylor & Francis Group, LLC

Das könnte Ihnen auch gefallen