Ok, I think this refers to the statistical confidence factor (having enough samples to assume data are representative of a larger population).

It's difficult to tell with the elements I have, but I think we face two different notions here:

1) statistical data

some marketting survey yelds results such as "40% of the potential customers like pink bananas". Such numbers are statistical data.

Now if there were only 5 answers to this question (do you like pink bananas?), the number of answers would not be high enough to have confidence in the result (if you see only 5 cars at a crossroad and 2 of them are blue, it is not enough to say 40% of the cars in France are blue).

2) confidence factor and number of samples

this number measures the confidence you can have in a statistical measure. Basically, the more answers to a survey, the higher the confidence factor. In the case of our bananas, the confidence factor would be so low (let's say 10%) that the statistic would be considered meaningless.

For instance you could say "the surveys show that 40% of the customers like pink bananas with a confidence factor of 10%".

Now take a deep breath. Here the aim is to decide whether you will try to sell rather pink or green bananas. So you want to compare statistics. (40% like pink bananas, 60% like green bananas, let's try to sell more green bananas).

But before trying to do such comparisons, you have to analyze the raw survey data to prune the non-significant results. The first thing you will do is reject data with a low connfidence factor (in your case, anything < 95%). Since other more subtle factors may cause the confidence factor to be very high even though the number of samples is not sufficient to be sure the information is significant, it is customary to also reject data based on too small a number of samples (in your case, anything based on less than 30 samples).

So you will say "40% - 60%" is a 20% difference, that is significant. But I made sure the informations I just compared had a sufficient confidence factor (> 95%) and a sufficient number of samples (> 30) (or else the comparison would have been pointless).

Finally, with this in mind, I propose "the analysis retained only significant differences between survey data with a confidence factor > 95% and based on at least 30 samples"

LOL that made me think hard remembering my math classes, but I hope that covered the issue

.