Data analysis for tetrad discrimination tests

To use this part of the toolkit, you will need the results of your tetrad test. Specifically, you will need:

Your prior intuition for how easy it would be for panelists to tell the difference between the samples. You wrote this down, right?
The number of subjects who participated, which we can call \(N\).
The number of subjects who got the right answer, which we can call \(x\). By necessity, \(x≤N\).

Once you’ve got those, we’re ready to get analyzing!

What you’ll get from this data analysis

Before we get to the analysis, a quick, intuitive explanation. Above, we talked about two important numbers: \(x\) and \(N\). In fact, the proportion of correct answers, \(x/N\) is the key quantity we’re interested in here. Specifically, we’re interested in what insight \(x/N\) gives us into the plausible proportion of people noticing a difference in your two products, A and B, which we’ll call \(\pi\) (\(\pi=p=\) proportion). Because \(\pi\) is a proportion \(0≤\pi≤1\), but we will present results in terms of both \(\pi\) and in terms of the number of people in 100 who would notice a difference.

Intuitively, if \(\pi\) is closer to \(1\), then most people will notice a difference. If \(\pi\) is closer to \(0\), then very few people will. Behind the scenes, we’re using \(x/N\), combined with your prior estimate of how likely you think there is to be a sensory difference between A and B, to calculate a plausible range of \(\pi\) values that could lead to that \(x/N\).

By giving some worst- and best-case outcomes, we’ll tell you what you should expect to see, based on your prior intuition and your actual data, if you brought your products to market.

What you’ll get from this data analysis

Further Reading