In short
Each sensory test answers one question. Use a triangle test to ask "are these different?", a paired comparison for "which has more of this attribute?", ranking for "what order are these in?", descriptive profiling for "how exactly do they differ?", and a preference test with real consumers for "which do people like?". Using the wrong test is one of the most common errors in evaluation.
What you will learn
- Match a question to a sensory test
- Calculate guessing probability for common tests
- Avoid the common mismatches
Before this lesson: Design a Wear Test: Study Guide
Choose by the question
| Question | Test | Chance of a correct guess | Who |
|---|---|---|---|
| Is there any perceptible difference? | Triangle | 1 in 3 | Screened assessors |
| Same as this reference? | Duo-trio | 1 in 2 | Screened assessors |
| Which is stronger, or sweeter? | Paired comparison | 1 in 2 | Assessors who agree on the word |
| What is the order of intensity? | Ranking | Depends on the number of samples | Assessors |
| How exactly do they differ? | Descriptive profile | Not applicable | A trained panel |
| Which do people prefer? | Preference or hedonic | Not applicable | Target consumers |
The reference guides on the triangle test and consumer testing go further into each. Formal versions exist in standards such as ISO 4120, and our home versions are not those.
Chance levels, worked
You saw the triangle arithmetic in blind practice. A paired test has a higher chance level, so it takes more correct answers to mean the same thing:
| Correct | Paired or duo-trio (1 in 2) | Triangle (1 in 3) |
|---|---|---|
| 5 or more | 62% | 21% |
| 6 or more | 38% | 7.7% |
| 7 or more | 17% | 2.0% |
| 8 or more | 5.5% | 0.34% |
| 9 or more | 1.1% | 0.035% |
So 7 of 10 is respectable for a triangle, and weak for a paired test. These figures assume independent trials, which repeated sniffing by one person does not fully meet, so treat them as approximate.
Common mismatches
- "Which do you prefer?" to find out if two are different. People give an answer even if they cannot tell them apart.
- A triangle test to decide which is better. It asks only if they differ.
- A paired test with an undefined word. Everyone needs to mean the same by "fresh".
- Asking trained assessors what consumers will like. Their job is to detect and describe.
- Asking friends for preference and then reporting it as market data. They are not your market, and they want to please you.
Keeping it fair
- Code samples with random numbers, and vary the order between people.
- Prepare samples at the same time, on the same substrate and at the same strength.
- Limit the number of samples, and allow rest.
- Separate assessors, and do not let them discuss it.
- Decide what result would change your mind before you look.
Practical exercise: Match tests to questions, then run one
- Match each to a test, with a reason: (a) did a new lot change our product? (b) which of two mods should we show the client? (c) is B sweeter than A? (d) how does B differ from A? (e) can a cheaper version pass as the original?
- Choose one of (a), (c) or (e), and run ten trials with one or more people.
- Compare your score with the table above.
- Write your conclusion, including what you did not test.
Suggested answers: (a) triangle; (b) a preference test with the client's target people, plus judgment against the brief; (c) paired; (d) descriptive; (e) triangle, followed by consumer testing, knowing that failing to detect a difference is not proof of none.
You have done this well if your conclusion uses the right chance level, and does not claim more than the test asks.
Common mistakes
- Treating 6 of 10 as success. Check the table for your test.
- Reporting "no difference" as "identical". It means none was detected.
- Letting one strong opinion spread. Do not discuss samples until all answers are in.
Key terms
- Paired comparison
- A test showing two samples and asking which has more of a named attribute. Guessing is right half the time.
- Duo-trio test
- A test with a known reference and two coded samples, one matching the reference. Guessing is right half the time.
Knowledge check
Written and reviewed by the RUŌOD Lab team. This is general education, not legal, regulatory, medical or safety advice. Example dilutions and exercises are for learning, are not validated products, and have not been safety-assessed. Completing a lesson confers no qualification. How we check these lessons.