Why is this test done?
For most of the food companies, researching and knowing consumer perceptions on their food/brand is extremely critical.
For example, you own a brand of food product. Suppose, due to some new constraints, like maybe discontinuance of raw materials, or any hindrance, you need to change the recipe or make subtle changes to the product.
It puts the brand at risk, due to the familiarity of the brand with the consumers. Read Coca Cola case of making changes to their product here. So, it becomes very important to test it out and see that consumers perceive no change in the product. That is where Triangle Test comes in.
Triangle test is one of the simplest tools to be able to ascertain if a set of people can differentiate in the given samples. It is a type of discrimination test based on sensory observations.
This test can be used to determine if the new formulation or ingredient change has any perceivable effect on the food.
Typically, a panel is formed. Each person in the panel is given 3 samples to taste, out of which 2 are identical. However, there is no indication given as to which samples are identical. The panel then needs to identify the odd sample. If they are not able to differentiate, then the changes made to the product are not perceivable.
Advantages of this methodology:
- Very simple to execute
- Gives very clear direction if products are perceivable different or not
- People can have biases towards the first sample they taste. So it is always recommended to rotate the samples across panelists. For example, for 2 samples A and B, possible combinations are: AAB, ABA, BAA, BBA, BAB, ABB
- Each panelist should keep the views private and record on their sheet. They should not give any verbal/ non-verbal cues for any of the samples.
- Each sample should have a 3 character name. The name should not induce any bias whatsoever
- As much as possible, the products should not be visually very different. In case they are, please ensure enough measures are taken to remove that bias
- This test should not be used to gauge preference. The purpose of this test is to gauge difference only
A Chi-Squared distribution can be used for evaluation of the Triangle Test with a large sample size.
O = Observed correct answers
E = Expected correct answers
To obtain the number of expected correct answers, multiply the number of panelists (n) with the probability of getting a correct answer.
In Triangle tests, this probability is considered to be 1/3. Thus the probability of incorrect answers is 2/3.
Alternatively, sometimes the results are clear enough and we might not need statistics for determining. For example, lets look at the case below.
Illustration with an Example
Lets assume Nestle wants to bring about changes to its product, Maggi Instant Noodles, due to the controversies in the recent years regarding its ingredients. Now, Maggi is a tremendously popular brand and many youngsters today hold it close to their hearts as the food they loved as a kid.
So, if Maggi wants to make changes in its recipe, it can’t risk letting its fans down with a variance in taste. In this case, Maggi should use a Triangle Test to ensure consistency in taste.
After the recipe is finalized, Maggi would administer the Triangle test to a set of panelists. Lets say, for the sake of this example, they used 24 members (n).
The members were given a questionnaire and the samples were handed in a random manner. Each respondent would receive a combination of the samples to taste. After tasting, they would try to choose which samples are different.
After going through the responses of the 24 participants, lets say, there were 8 correct ones and 16 incorrect ones. Thus, its 1/3 of the entire lot (as mentioned before in terms of probability). It is very clear here that 1/3 people are getting it wrong and are unable to perceive any difference. So, Maggi can go scot free with the change to its recipe.
The second possibility is when there are many correct responses. Lets say, 18 of them are able to comprehend a difference in taste. This spells trouble. It means that the R&D team has a lot more ground to cover in order to match the taste.
The last possibility for the test is when it gets complicated (statistically). If there exists a case where the numbers get very close, i.e., almost equal number of people are correct and incorrect, then we might have to resort to statistical tools to understand it better.