Our recommended sample size is 300 ‘recorded’ sessions, which guarantees enough 'delivered' sessions for robust, repeatable results. While smaller sample sizes are useful for detecting trends, the results are less 'certain' and repeatable.
There is a level of statistical variability that is inherent in the results, which is affected by the total sample size. As sample sizes increase, the sample variability decreases, and we have greater precision.
The graph found here illustrates the statistical variability inherent in the results. The graph maps the chances of generating an accurate result* (+-4%) for sample sizes from 50 to 2000. The probabilities illustrated by the graph apply to all emotion metrics.
Examining the graph, you’ll see that if the 'Real Probability' % of people showing Happy, for example, is 15%, the statistical probability that a sample of 50 would also show Happy of 15% (+- 4%) is 57%.
If you increased the sample size to 150, the chances that the % of people showing Happy would be 15% (+- 4%) is 83%.
By using a sample size of 300, we are able to keep the chances of getting an accurate estimation between 75%-100%.
* Accurate result = the exact result that you’d get from a much larger sample with the exact same specifications