SENSORY EVALUATION: DISCRIMINATION TESTING
Diagram
Two ways to ask: are these different
Triangle test
Three coded samples are served: two are the same and one is different. The panellist picks the odd one out.
↓
Duo-trio test
A reference sample is served first, then two coded samples: one matches the reference and one does not. The panellist picks the one that matches.
↓
Same question
Both ask only one thing: can the panellist tell the products apart. Neither asks which is liked or how they differ.
Figure 1: The triangle test uses three samples with no reference; the duo-trio test uses a reference plus two. Drawn in our own style from the discrimination testing chapter (p. 116) and Types of Discrimination Tests (p. 117) of Lawless and Heymann, Sensory Evaluation of Food.
Introduction
The triangle test and the duo-trio test are the two most used discrimination tests, and both ask panellists to find a difference, but they serve the samples differently. In the triangle test, three coded samples are served, two the same and one different, and the panellist picks the odd one out. In the duo-trio test, a reference sample is served first, then two coded samples, one matching the reference and one not, and the panellist picks the one that matches. Lawless and Heymann cover both in the discrimination testing chapter (p. 116) and in Types of Discrimination Tests (p. 117).
The core difference is the reference: the triangle test gives none, and the duo-trio test gives one, which changes how the panellist works and how likely a guess is to be right.
Quick comparison
Table 1: Triangle test compared with Duo-trio test
| Aspect | Triangle test | Duo-trio test |
|---|---|---|
| Samples served | Three coded samples | One reference, then two coded samples |
| Reference sample | None | Yes, served first and labelled as the reference |
| Task | Pick the odd one out | Pick the sample that matches the reference |
| Chance of guessing right | One in three | One in two |
| Memory load | High: the panellist must hold three samples in mind | Lower: the reference is tasted first and can be re-tasted |
| Best use | When the difference is small and the panellist must work hard to find it | When a reference product exists and the question is whether a new sample matches it |
| Sensitivity | Usually more sensitive, because guessing right is harder | Usually less sensitive, because guessing right is easier |
| Panellists needed | Fewer, because each answer carries more weight | More, because each answer carries less weight |
| Tasting order | All three are tasted in a set order | The reference is tasted first, then the two coded samples |
| Result | A yes or no: the products differ, or no difference was found | A yes or no: the sample matches the reference, or it does not |
Scientific explanation
Definition
The triangle test is a discrimination test in which three coded samples are served, two the same and one different, and the panellist must pick the odd one out. The duo-trio test is a discrimination test in which a reference sample is served first, then two coded samples, one the same as the reference and one different, and the panellist must pick the one that matches the reference. Both are described in the Types of Discrimination Tests section (p. 117) of Lawless and Heymann.
Both are discrimination tests, which means they ask only whether a difference can be sensed. Neither asks which sample is liked, or what the difference is.
Working principle
The triangle test works by forcing a choice among three: the panellist tastes all three, compares them in memory, and picks the one that stands apart. Because there are three samples and only one is odd, a panellist guessing at random is right one time in three.
The duo-trio test works by matching to a reference: the panellist tastes the reference, holds its taste in mind, then tastes two samples and picks the one that is the same as the reference. Because there are two choices, a panellist guessing at random is right one time in two, which is why the duo-trio test needs more panellists to reach the same certainty.
Apparatus
Both tests need a sensory test room with separate booths, coded cups or plates, water and plain crackers to cleanse the palate, and a ballot form where the panellist marks the answer. The triangle test needs three coded samples per panellist, and the duo-trio test needs a reference plus two coded samples per panellist.
Measurement procedure
For the triangle test, the panellist is served three coded samples in a set order, tastes them, and marks the odd one out. For the duo-trio test, the panellist is served the reference first, tastes it, then tastes the two coded samples and marks the one that matches the reference. In both, the order of the samples is balanced across panellists, so no sample gains from being served first or last.
The answers are counted, and the count is checked against the binomial probabilities for guessing: the books give binomial probability tables for discrimination tests in an appendix (p. 795). If far more panellists are right than guessing would give, the products are judged different.
Factors affecting readings
Fatigue matters: tasting three samples is tiring, so the samples must be small and the palate cleansed between them. Carryover matters too: a strong flavour from one sample can mask the next, so the order is balanced and the panellist rinses between samples. The size of the real difference matters most: a large difference is found by almost everyone, and a tiny difference by almost no one.
Formulas and calculations
The guessing probability follows from the design of each test. In the triangle test there are three samples and one is odd, so the chance of guessing right is 1 in 3. In the duo-trio test there are two coded samples and one matches the reference, so the chance of guessing right is 1 in 2.
Worked example: if 30 panellists take the triangle test and guess at random, the number expected right by guessing alone is 30 x 1/3 = 10. If 30 panellists take the duo-trio test and guess at random, the number expected right is 30 x 1/2 = 15. The books give binomial probability tables for discrimination tests in an appendix (p. 795), and the count of right answers is checked against those tables to judge whether the products truly differ.
Industrial applications
- The triangle test is used to check whether a change in recipe, supplier or process has changed the product.
- The triangle test is used to check whether a cheaper ingredient can replace a dearer one without being noticed.
- The duo-trio test is used to check whether a new sample matches a reference product, such as a standard or a competitor.
- The duo-trio test is used in quality control, where each batch is matched against the reference.
- Both are used before a product is launched, to prove that a change has, or has not, changed what the buyer tastes.
Advantages and limitations
- Triangle test, advantage: more sensitive, because guessing right is harder, so fewer panellists are needed.
- Triangle test, limitation: harder for the panellist, because three samples must be held in mind, and fatigue can spoil the result.
- Duo-trio test, advantage: easier for the panellist, because the reference is there to compare against.
- Duo-trio test, limitation: less sensitive, because guessing right is easier, so more panellists are needed for the same certainty.
FAQ
Which test is better, triangle or duo-trio?
Neither is better for all uses: the triangle test is more sensitive and needs fewer panellists, while the duo-trio test is easier for the panellist and suits matching a sample to a reference. The question decides the test.
Why does the triangle test need fewer panellists?
Because guessing right is harder: one in three in the triangle test against one in two in the duo-trio test. Each right answer in the triangle test is stronger evidence, so fewer answers are needed.
Do the panellists know which sample is which?
No: the samples are coded, and the panellist does not know which is which. In the duo-trio test the reference is labelled as the reference, but the two coded samples are not.
Can these tests tell me which product people like?
No: they only tell whether a difference can be sensed. Liking is asked in hedonic tests, not in discrimination tests.
What if nobody can tell the products apart?
That is a result too: it means no difference was found by the panel, which may mean the products are the same to the senses, or that the difference is too small for the panel to find.
References and further reading
- Harry T. Lawless and Hildegarde Heymann, Sensory Evaluation of Food: Principles and Practices, Springer, 1999 (Google Books page read)
- Jian Bi, Sensory Discrimination Tests and Measurements, 2nd ed. Wiley, 2015
