1.

Discrimination test ( Nhóm phép thử phân biệt)

Definition: Discrimination tests are used to determine whether two products physically
different can be considered as perceptually different
Purpose:
- Determine if there is a difference between two products
- Use two products with very small, imperceptible differences in one or more sensory
properties
Applications:
- Screening and training testers
- Determination of sensory thresholds
- Evaluation of defects
- Quality assurance/quality control (QA/QC)
- Evaluation of effectiveness when changing components or manufacturing processes (eg
to reduce costs or change suppliers)
- Preliminary assessment
Phân loại:
- Nhóm phép thử phân biệt tổng thể: Simple difference: “which sample is different?”
+ Triangle test
+ A not A test
+ Duo – Trio test
+ Paired comparison tests
+ Two out of live
+ Tetrad
- Nhóm phép thử phân biệt thuộc tính: Directional difference: “which sample is sweeter?”
+ Two alternative forced choice (2AFC)
+ Three alternative forced choice (3AFC)
Phép thử phân biệt tổng thể: Simple difference
Goal: To determine whether an overall difference exists, where no specific attributes
can be identified as having been affected.
Applications:
- Method applicable when the nature of the difference is unknown
- Applicable to homogeneous products
- Doesn’t determine neither the size nor direction of difference between samples.
TRIANGLE TEST
Purpose:
- Identify the difference between two products without knowing the nature of the
difference.
- This test is used in cases where the difference between two products is relatively small
Principle:
- The assessors receive a set of three samples and are informed that two of the samples
are similar and that one is different.
- The assessors report which sample they believe to different, even if the selection is
based only on a guess.
- QUESTION: Which sample is different from the two other?
How to present the samples?
- Random and balanced.
- Number of assessors: 20 – 40
- The probability that the tester chooses the correct answer at random is 1/3 .
How to analyze the data?
- Perform a statistical test
DUO – TRIO TEST
Purpose:
- Determine the overall difference between two products regardless of their sensory
properties.
- Model R is the model the tester is familiar with, the other is a new product
Principle:
- The tester will be given one control sample R (Reference) and two other samples, the
tester knows in advance that one of these two samples is the same as the control
sample.
- The tester tests sample R several times to "belong to the pattern", then tests the
remaining 2 samples and determines which one is the same as the control sample.
- The probability of the correct answer is 1⁄2.
How to present the samples?
- Random and balanced
- Number of assessors: more than 15 people or better than 30 people. Need not be
experts,
A NOT A
Purpose:
- Determines whether a product range is identical to a reference material. So this is
essentially a series of paired-difference tests. The results of this test indicate whether
the member can meaningfully distinguish between the samples, when the samples do
not appear simultaneously.
Principle:
- Assume “A”: reference sample, “not A”: product. First, the tester must be "learned" to
recognize sample A, then this sample is taken. Next, the tester will try a sequence of
coded samples including sample A and sample not A. The tester must determine which
sample is A and which is not A.
- Since testers cannot test two samples at the same time, they must remember, compare
the two samples and decide if they are the same or different.
- This test is one-sided when the experimenter knows the exact answer to the question
posed to the tester, that is, knows in advance whether the two samples are similar or
different.
How to present the samples?
- Balanced and random, the number of times a occurs is equal to the number of times Not
A occurs
- Number of assessors: 50
TWO OUT OF FIVE
Principle: The assessor is presented with a set of five coded samples and is told that
two are of one type and three of another. The assessors are asked to group the two sets
of samples.
QUESTION: Sort the products into two groups of 2 and 3 products.
How to present the samples?
- 20 possible orders of two products A and B.
- Present each order and equivalent number of time.
How to analyze the data?
- Sort the products into two groups of 2 and 3
TETRAD TEST
Principle: The assessor is presented with a set of four coded samples and is asked to
group the samples in two groups of two samples based on their similarity.
QUESTION: Sort the products into two groups of 2.
How to present the samples?
- 6 possible orders of two products A and B.
- Present each order and equivalent number of time.
How to analyze the data?
- Sort the products into two groups of 2.
Phép thử phân biệt thuộc tính: : Directional difference
Goal: To determines whether or not there is a perceivable difference in the intensity of a
specified attribute between two samples.
Applications:
- Method applicable when the nature of the difference is known.
- More sensitive than simple difference.
2AFC
Purpose:
- Compare pairs of samples prepared from two products that differ in a particular
organoleptic property.
- The sensory properties are indicated at the outset, so this test has a fairly high
sensitivity.
- If there are more than two samples, we use the analysis of variance (ANOVA) method.
Principle:
- The assessors receive two samples.
- They report which sample the believe is more (attribute) than the other, even if
selection is based only on a guess.
- The probability of getting the correct answer randomly in this test is 1/2
- QUESTION: Which samples is sweeter?
How to present the samples?
- Balanced and random.
- Number of assessors: min 30 assessors.
How to analyze the data?
- One sided paired comparision test.
- Two sided paired comparision test.
3AFC
Purpose:
- Used when we want to distinguish differences in a definite sensory property.
Principle:
- The assessors receive a set of three samples.
- The assessors report which sample they believe is more ( attribute) than the others,
even if the selection is based only on a guess.
- QUESTION: Which samples is sweeter?
How to present the samples?
- Random and balanced, only 3 sample arrangements.
- Number of assessors: min 24 assessors.
How to analyze the data?
2. Descriptive tests (Nhóm phép thử phân tích mô tả) – cấu trúc mùi vị
Goal: To describe the sensory characteristics of a product, and to use these
characteristics to quantify differences between products
Contrary to informal product tasting, descriptive analysis limits:
- Bias
- Subjectivity
- Poor control of variables
- Using incorrect assessors
- Patchy information
Applications
+ To describe current products in the market ( mapping a market) to develop a new
products from gaps in existing market map.
+ To corelate sensory and instrumental/consumer data.
What sort of panel is required?
- Trained panel: more able to describe the subtle differences between samples
- Two types of trained panels:
+ Interal: within the company
+ External: outside the company
How to do descriptive analysis?
- Compile a list of descriptors that include all the sensory characteristics of the products.
- Evaluate the intensity of the perceived sensation for each descriptor.
- Set up the product profiles by using all the descriptors (spider graph, histogram,…)
- Compare the profiles of the different product ( visually + statistically).
Four main steps of descriptive analysis
STEP 1: Panel selection
- Ability to:
+ Express oneself
+ Understand and answer a questionnaire
+ Discriminate, identify, memorize stimuli
- Interest in participation and availability
- General good healthy
 Prescreening questionnaires
Acuity tests
Ranking/rating tests
Identification tests
Scaling ability tests
Screening questionnaire
Time:
- Are they any weekdays that you are not available on a regular basis?
- How many weeks of vavation do you plan to take this year?
Healthy:
- Do you have any of the following?
 + Dentures
+ Diabeters
+ Oral or gum disease
+ Hypoglycemia
+ Food allergies
+ Hypertension
- Do you take any medications which affect your senses, especially taste and smell?
Motivation:
- Why would you like to participate in this panel?
- When you see a new food in store, do you want to buy it and why? Why not?
- What is your favorite dish and why?
- When you go to a restaurant, do you prefer choosing a familiar dish or taste dishes you
don’t know?
- What would you use to replace soy sauce in a recipe and why?
- Descibe some of the flavors found in a pho?
Screening tests
- Ability to use a scale
- Name odor matching
- Triangle test: Kro vs Heineken
- Identification of tastes and sensations
- Ranking of sweet perception in water
- Verbal fluency
STEP 2: Generation of a lexicon
Goal: Decribe the product category completely without overlapping
Procedure:
- Word generation
- Reduction of the list into a working list
+ Rearrange the list in family of terms ( appearance, aroma, textures…)
+ Discuss with panelists to suppress hedonic, redundant, non relevant terms
- Choice of referents and definitions for the descriptors
The words should describle ONE unique sensory characteristic.
They should be:
- Simple, unambiguous, understandable to facilitate communication and broad enough to
include all the perceived sensations.
- Analytic (non hedonic), relevant, discriminant, discriminant and independent
2 approaches:
- Use a predefined list: Time saving, comparison of results between different panles
- Create the list with the panle: better appropriation of the attribites by judges
Assessors are exposed to a set of samples that represent the family of products
They are asked to describe individually their perception of the samples ( aspect, odor,
taste & aroma and texture) using familiar vocabulary.
Reduction of the list: step 1
- Selection of terms cited mỏe than once per product.
- Deletion of hedonic terms, quantitative (too much, little,…) and irrelevant.
Reduction of the list: step 2
- List from the 1St sorting
- Same products
- Intensity rating ranging from o to 5 for each attribute.
Reduction of the list: step 3
- Principal component analysis (PCA) and Hierarchical cluster analysis (HCA)
- Group synonyms or antonyms together: correlation matrix + dendrogram
STEP 3: Panel training
High variability between panelists:
- Vocabulary
- Sensitivity
- Use os the scale
Training objective (mục đích): qualitative and quantitative alignment of tasters
 Need to be exposed to a wide range of products
Need to be familiarized with the use of the sacle
Need to receive feed back to rectify any problem
Qualitative training: All tasters must use the same term to describe the same sensation
( ví dụ phân tích mùi vị, cấu trúc)
- Memorize odors, tastes, aromas, textures
- Make the tasters discuss by tasting the products again
- Define precisely the descriptors
- Set up evaluation protocols (especially for texture)
Quantitative training: All tasters must be use the same term to describe the same
sensation ( ví dụ phân tích mô tả định lượng)
- Choose quantitative references
- Calibrate over know concentration ranges to judge intensities
- Train the tasters to use scales ( entire length of the scale)
How to set up the training? For each attribute
- Ranking tests
 Panelists receive 4 samples covering the range of didferences likely to be encountered in
the product category and asked to rank them.
- Rating tests
Panelists receive 4 samples and are asked to score them on an intesity sacle
Start with large differences, then collect samples which represent smaller differences
within the product category
Encourage panelists to refine the procedures for evaluation and terminology with
definitions and references.
How to set up the training? Final pratice
- Should approach the testing situation in which the panel will be used:
Same type of products
Same experimental conditions
Several replicates
- Panelists should receive feed back on their performance compare to that of the panel
STEP 4: Product rating
Determine the specific properties of the product:
- Each member of the panel will evaluate the samples (products) in turn.
- Each evaluation is a sample and scores all the sensory properties of that sample.
- The experiment must be repeated, perform at least two test sessions (or 2 repetitions).
Data processing method
- Analysis of variance (Anova - analysis of variance)
+ Evaluate the consensus of the panel
+ Evaluate the difference between products
- Analyze the main part
+ Grouping the evaluation products based on sensory properties of the product.
3. Consumer test ( Nhóm phép thử thị hiếu)
Purpose:
- Assessment of curent and potential customers’hedonic respones to a product or
specific product characteristics
- Ít measures preferences for products or magnitude of like/dislike for a product
- Products can be evaluated on:
Texture ( mouth feel)
Taste or flavor
Appearance or odor
Smell or aroma
Hearing
Tính chất - Nature
- Hedonic reactions depend on personal experience
- Not stable
- Large inter individual variability
 Work with a large number (N>100, as a rough rule of thumb) of unstrained individuals
who consume the producr you are interested in.
Key points in designing good consumer tests – nguyên tắc thiết kế tốt p,thử thị hiếu
a. Setting the right objective
The method chosen must fit the objectives of the research question so:
- Be clear on test objectives: Think of possible test outcomes and identify appropriate
actions
- Do one test at a time to avoid confusion
b. Asking the right people
Participants must be chosen carefully to ensure that the results will generalize to the
population of interest:
- Frequent users of your product or similar product
- Correspond to the targeted population
- Avoid trained panelists or persons having knowledge about the product (companies or
laboratory personnel)
c. Asking the right questions
Consumers might perceive product as a whole pattern, rather than in an analytical way,
have very fuzzy concepts about specific attributes and might have difficulties making
absolute judgments so:
- Avoid asking consumers about specific analytical attributes, use global questions
- Use comparative judgments rather than absolute one
How are consumers recruited for studies?
Central location test (CLT)
- Consumers are recruited and invited to a testing facility to participate in the test:
+ Advantage: Test administrators have control over the test design, product preparation
and presentation.
+ Inconvenient: Situation very different from real life, so problems of extrapolation.
Hone use test (HUT)
- Consumers are cruited to take products home and use them according to instructions.
They fill out a questionnaire and return it.
+ Advantage: more ecological than CLT tests
+ Inconvenient: less control that CLT tests, more expensive, more time consuming.
How to decide which test using?
The first rule in designing a consumer test is to keep it easy and understandable by all.
 Four main approaches:
- Preference/choice (pair comparison tests)
- Preference ranking
- Rated acceptability (category or line sacles)
- Just – about – right (JAR) scale
PAIRED – PREFERENCR TEST (Phép thử ưu tiên cặp đôi)
Principle: A pair of coded samples is presented for liking comparison
Which one of the two coded samples do you prefer?
Presentation: Balance A/B B/a and radom
Results: Count nuber of respones A et B
 Chi2 or binomial test
RANKING TEST (Phép thử so hàng thị hiếu)
Purpose:
- Use when in doubt products differ in a property
- This test can also be used for the purpose of determining the tester's preference for a
product of the same type.
Principle: A series of coded samples is presented for ranking
Rank samples according to your preference
Presentation: Balance serving order to avoid position and carry over effects. Number of
assessors: 16 people or more, at least 8
Results: Compute the sum of the rank for each product
 Friedman test
RATING SCALES ( Phép thử cho điểm thị hiếu)
Purpose:
- Determines how much there is a difference in some sensory property between two or
more products. The logic of this test í to quantify the percieved intensity of a sensory
property.
- This test is used only when the products to be evaluated are different and we have
doubts about the properties with which they differ.
Principle: A series of coded samples is presented for rating
Indicate the degree to which you like or dislike the samples
Presentation: Balance serving order to avoid position and carry over effects.
Results: Compute average scores and standard deviations or transform in ranks
 Anova or Friedman test
Three main catogories of liking scales
- Category scales: Verbal anchors are ascociated with some or all the categoties
- Numberical sacles: Numberiacl anchors are associated with sơm or all the categories
- Line scales: Verbal anchors are positioned at the left and right end of a line
 Category scales are the most broadly used
9 point hedonic scale
- Like extremely - Dislike slightly
- Like very much - Dislike moderately
- Like moderately - Dislike very much
- Like slightly - Dislike extremely
- Neither like nor dislike
Main inconvenient
- Avoidance extreme points
- Does not translate well
Some people suggest using other scales: 5,7, or 11 point scales or facial hedonic scales
TCVN 3215 – 79
- Overall assessment of the product quality level in comparison with the standard or a
product of the same type on all sensory criteria: color, odor, taste and state.
- Since the criteria play a different role in the overall quality of the product, the score
values for each criterion are multiplied by a value called the important coefficient.
- When assessing sensory quality by a panel, the quality score of a given indicator is the
average score of the members multiplied by its important coefficient.
- The total score of the criteria is the quality score of the product
- Check all sensory criteria or each individual sensory criteria (state, color, odor, taste) of
each type of product and goods.
- Using a score system of 20 built on a unified scale of 6 steps of 5 points (0 to 5), in
which 0 points correspond to defective quality, and scores from 1 to 5 correspond to
decreasing defect levels. At point 5, the product is considered to have no defects or
defects in the properties under consideration, and the product has good characteristics
that are clearly characteristic for a certain criterion.
- The sum of the important coefficients of all the indicators is equal to 4.
- The panel must consist of 5 to 12 experts with knowledge of the product being
evaluated. The council has a chairperson and a secretary to lead the council in its
working process.
- When evaluating members working independently, assign scores to a form and submit it
to the secretary after hours.
- The committee secretary will summarize the points of the members and from there
calculate the quality score of the product.
- If a sensory indicator gets a score of 0, then testing should be done again for that
indicator.
- When the panel has decided to give a certain criterion a score of 0, the product is
evaluated with a total score of 0.
- For a homogeneous product sample, a panel member's comment is rejected when it
differs by more than 1.5 points from the unimportant mean score.
Quality level Common GPA required without weight
point
Best 18,6-20,0 The most important indicators≥4.7
Good 15,2-18,5 The most important indicators≥3.8
Medium type 11,2-15,1 Each indicator≥2.8
Poor Grade (still available) 7,2-11,1 Each indicator≥1.8
Very poor type (not for sale) 4-7,1 Each indicator 1.0
Damaged type 0-3,9

1. Một công ty sản xuất nước trái cây đang xem xét chuyển đổi công thức nước ép trái
cây mới. Công ty quyết định thực hiện thử nghiệm so sánh theo cặp (2-AFC) với mục
tiêu xác định xem có sự khác biệt đáng kể về độ chua giữa hai mẫu: mẫu A (công thức
mới có bổ sung Vitamin C 3mg/l) và mẫu B (công thức cũ công thức) với mức ý nghĩa
1%. Một nhóm gồm 39 người đánh giá đã tham gia vào một thử nghiệm so sánh theo
cặp. Kết quả cho thấy 30/39 người đánh giá chọn mẫu A chua hơn mẫu B.
A juice products company was considering switching the new recipe’s fruit juice. The
company decided to run a paired comparison test (2-AFC) with an objective to
determine if a significant difference existed in sourness between the two samples:
sample A (new recipe with addition of Vitamin C 3mg/l) and sample B (old recipe) at
1% significance level. A panel of 39 assessors participated in a paired comparison test.
The results showed that 30 out of the 39 assessors selected sample A as sourer than
sample B.
According to the results: 30 out of the 39 assessors selected sample A as sourer than
sample B (OR: A panel of assessors: 39, 30 ssessors selected sample A as sourer than
sample B)
From the table in Appendix 4 - Critical values table for paired comparison test for
difference (one tailed), for a panel of 39, the minimum number of responses required to
determine that a difference exists at 1% significance level is 28 (OR: the minimum
number of correct responses to reject the null hypothesis of ‘no difference’ at 1%
significance level with a total number of 39 assessors is 28).
Because 30>28 (the test result DOES NOT exceed this value): There is a significant
(INSIGNIFICANT) difference in sourness between the two samples of fruit juices in which
sample A, the new recipe product, is significantly (INSIGNIFICANTLY) sourer than
sample B, the old recipe product.
Conclusion: The sample A was sourer than sample B (at 1% significance level). (THE
PANELISTS CAN NOT DISTINGUISH THE DIFFERENCE IN SOURNESS BETWEEN SAMPLE
2. Một công ty giải khát muốn biết sản phẩm của công ty đứng ở vị trí nào trên thị
trường. Công ty quyết định tiến hành phép thử so hàng để xác định xem liệu có sự
khác biệt có ý nghĩa về mức độ yêu thích đối với 4 sản phẩm nước giải khát đứng đầu
trên thị trường hay không? Hội đồng gồm 10 người tiêu dùng tham gia đánh giá 4 sản
phẩm P1, P2, P3, P4 (P1: là sản phẩm của công ty). Kết quả đánh giá cảm quan thu
được như sau:
Người thử P1 P2 P3 P4
1 2 4 1 3
2 2 3 1 4
3 2 4 1 3
4 3 4 1 2
5 2 4 1 3
6 2 3 1 4
7 2 4 1 3
8 3 4 2 1
9 4 2 3 1
10 3 1 2 4
Các sản phẩm được xếp hạng từ 1 -> 4 theo thứ tự từ ít yêu thích nhất đến được yêu
thích nhất. Chọn α=5%, tương ứng z=1,96 được lấy từ phân bố chuẩn 2 đôi với α=5%.
Với kết quả thu được ta kết luận gì?

According to the results:

Product P1 P2 P3 P4
Rank sum 25 33 14 28
The Ftest is caclutated as follows:
12
. ( R1 + R2 + R3 + R 4 ) −3 j. (p +1)
2 2 2 2
F test =
j . p . ( p +1 )
Where: j: no of assessor = 10
p: no of product = 4
Ri: the sum of rank
12
. ( 25 +33 +14 +28 ) −3.10 . ( 4+1 )=11, 64
2 2 2 2
 F test =
10.4 . ( 4+1 )
 From the table in Appendix 8, the critial value from the Friedman test is 7,82 ( with p-
1=3: degress of freedom, α=0,05)
Because the caculated F statistic exceeds this value (11,64>7,82)
 There is a significant differences in preference testing among samples with α = 5%
significant level.
Therefore, Fishers LSRD must be used to determine which samples are significantly
different in preference testing.

LSRD=z .
√ j . p . ( p+1 )
6 √
=1 ,96.
10.4 . ( 4+1 )
6
=11,316

Where z is taken tables for student’s distribution. Samples which rank sums differ by
more than 11, 316 are deemed to be significantly different. The results are
summarised in the following tables:
|P1-P2| |P1-P3| |P1-P4| |P2-P3| |P2-P4| |P3-P4|
8 11 3 19 5 14
Sample Ranking sum significant
P1 25 AB (III)
P2 33 A(I)
P3 14 B(IV)
P4 28 A(II)

Conclusion: the descent ranking order in preference between samples is as follows:

P2>P4>P1>P3