Web lists-archives.com

Statistics (was: FF losing popularity in so dramatic way?) -new thread

As always, the problem isn't the numbers, it's what they mean, how and why they changed, what they count, etc. there's a tendency to accept numbers uncritically, simply because they're numbers, which means some arithmetic was used to produce them. And arithmetic isn't a matter of opinion, right?

Google is pushing Chrome; a lot of free and commercial software includes Chrome, you have to be alert to prevent it being dumped on you during installation. Pop-ups urge you to switch. Etc. Thus a lot of Chrome is installed inadvertently.

"Most popular" is about percentages, not actual numbers. The difference between the top two items (browsers) may be only a percentage point or two. Generally speaking, a difference is less than one Standard Deviation for the sample may be "statistically significant", ie not a sampling fluke or error, but nevertheless not especially meaningful for a comparison.

Most important is the sample. Statcounter is visitor-tracking software. Thus its data is limited to that available via the websites of its customers. The data is unbiased if and only if the websites that use Statcounter are an unbiased sample of all available websites. In addition, the data must be weighted to allow for different rates of visiting, must count each visitor only once in each sample's time frame. IMO, it's unlikely that StatCounter's sample is unbiased.

An example of why counting visitors and browsers is difficult: For the last week and month, 98% of the visitors to my blog have used Firefox. Is that a good sample? No, because a) they are a self-selected group, unlikely to be representative of the total user population. And b) most of them are repeat visitors. Thus their browser will be over-represented in the count.

Wolf K.
"No keyboard present. Press F1 to continue."
general mailing list