5 December 2016 Christine URBANO Comments Off

Imagine you’ve found a new supplier whose flavors are less expensive, but before changing the ingredients in your top recipe, you want to make sure your customers won’t taste the difference, or at least that as few as possible will (e.g. fewer than 20%).

You might want to try the similarity triangle test. However, please note: while the protocol is the same as in a “classic” triangle test, this one is designed to show similarity, and therefore involves a different number of panelists and a different starting assumption.

First let’s consider the classic difference triangle test… where you only needed about twenty panelists to show a substantial difference between your products, with a risk lower than 5%. As a reminder, α risk corresponds to the assumption that your products are perceivably different although they are not. It is the risk you should take into account when you want to show a difference.

Now, if you’d like to show a similarity, the risk you should take into account is the assumption that your products are similar… although they are perceivably different. This is β risk. And this is where things get difficult: β risk does not depend directly on α risk, but on your assumptions in terms of “discriminator percentage” and on the number of panelists involved in the experiment.

For example – I will spare you the statistical details (even though I’m fond of it) – let’s say you want to show that at least 75% of your consumers will not perceive a difference between your old and new recipes: you will need a hundred or so panelists! That is way above the twenty or so people typically needed to show a difference… But that is the price to pay if you want to make sure your new recipe is indeed similar to the old one.

Here’s something worth knowing…! If you can substitute your triangle test with a two-out-of-five test, your testing power will always increase, while maintaining not only the same number of panelists, but also the same number of samples…! Thus, in the example above, 33 people would be sufficient to show a similarity…

Please do not hesitate to contact us if you need to set up your difference and/or similarity tests! We will be glad to assist you in striking the right balance between the number of panelists and the uncertainty in results. See you next month for a new focus on…