Discriminative report
The statistical verdict of the triangle, paired comparison (2-AFC), and duo-trio tests: how many correct answers it takes to confirm a difference.
What it is
The discriminative report is the result of the tests that answer a single question: can the panel tell one sample from another? It applies to three test types:
- Triangle (ISO 4120) — three samples; the panelist points out the odd one.
- Paired comparison (2-AFC) (ISO 5495) — two samples; the panelist points out which one meets a criterion.
- Duo-trio (ISO 10399) — one reference and two samples; the panelist points out which one matches the reference.
Unlike descriptive tasting, there is no spider chart or profile here: the output is a binomial verdict and the distribution of the panel's choices.
Where to find it
- URL:
https://app.areacacao.com/sensory-analysis/evaluations/<id> - How to get there: open the discriminative tasting under Evaluations and scroll down to the Results section.
Pro plan and above.
What the report shows
Summary line
The report opens with the test method, the number of responses from the panel, and the test's chance probability.
The probability of guessing correctly by pure chance depends on the test:
| Test | Chance |
|---|---|
| Triangle | 1 in 3 (33%) |
| Paired comparison 2-AFC | 1 in 2 (50%) |
| Duo-trio | 1 in 2 (50%) |
In the report's summary line, the app shows the method with the exact label Paired comparison (2-AFC).
Verdict
A highlighted box with the result of the binomial test on the panel's correct answers:
- Significant perceptible difference (p < 0.05) — the panel detected the difference beyond chance.
- Non-significant difference — the correct answers do not exceed what would be expected by chance; you cannot claim that the panel tells the samples apart.
Below it, you see the count of correct answers out of the total and the test values (p = ..., α = ...).
For a verdict to appear, you must have marked which sample was the odd one out when you created the tasting (the reference truth). If you did not mark it, the report shows only the distribution of choices, with the notice: "The statistical verdict requires marking which sample was the odd one out when the tasting was created."
Distribution of choices per sample
A bar chart showing how many times the panel chose each sample. The correct sample is highlighted in copper. The expected chance level (the percentage of correct answers you would get without telling anything apart) is shown in the chart caption, not as a line drawn over the bars. If the bar for the correct sample clearly exceeds that chance percentage, the panel is discriminating.
Literal chart caption:
Breakdown of the panel's choices. The expected chance line is X% correct by pure chance; the correct sample is highlighted in copper.
Effect size and power
The binomial verdict tells you whether there is a difference, but not how much there is, nor whether your panel is large enough to detect it reliably. For that, the report adds three measures:
| Measure | What it means |
|---|---|
| pd | Estimated proportion of the population that does tell the samples apart. It is obtained by correcting the observed correct answers for those expected by chance (Abbott's correction): it discounts those who got it right by luck and leaves only the real discriminators. It is the test's effect size. |
| pc | Observed proportion of correct answers: the panel's correct answers over the total responses, uncorrected. It is the raw figure the pd calculation starts from. |
| Recommended n | Panel size needed to reach 80% power, that is, to have an 80% chance of detecting a difference of the estimated magnitude if it really exists. |
If the total number of responses is smaller than the recommended n, the current panel may be insufficient: even if the verdict comes back "non-significant," the test did not have enough power to rule out the difference. The report warns you about this. Add more panelists or more replicates before concluding that the samples are indistinguishable.
How to interpret it
- A verdict of significant difference confirms that the change you tested (supplier, batch, technique) is noticeable.
- A verdict of non-significant difference does not prove the samples are identical: it means that, with this panel and this number of responses, the difference has not been detected. Increasing the panel size gives the test more power.