There are three kind of lies: Lies, damned lies and statistics.
— Benjamin Disraeli, British politician (1804 – 1881)
The thing about statistics and browser share has come up at Asa Dotzler’s blog, commented on at OperaWatch, and finally scowled at by Arve Bersvendsen at Virtuelvis.
So, what seems to be the problem? It starts with some problems figuring out how 60 million downloads of Opera 7 can translate into 1% browser share, while 50 million downloads of Firefox 1 translate into 8-10% browser share. That there were some versions of Opera starting with 7 does indeed explain quite a bit – people upgrade as new versions are available, even if it isn’t a full version number. However, not everyone download the browsers from sites which count the downloads: Some get it from CDs/DVDs on computer magazines, or download it from other download sites. Some download one copy but installs it on several computers. Trusting download stats is therefore fraught with insecurities when it comes to number of users.
Then there is the browser share of 1% – how accurate are the statistics? You may very well claim that the counters/services that provide the numbers for these stats are able to recognise Opera, even when masquerading as IE or Netscape/Mozilla, and you’d be right. Well – except when Opera 8 makes use of the two new, thoroughly camouflage modes. But the question of how accurate the services are remains, how good they are to recognise Opera every time – because Opera makes good use of the cache, and doesn’t show up as a new visit as often as other browsers. You can change the default settings for the cache in Opera, but if you don’t need to with your kind of browsing, why would you? Read more at Virtuelvis about the underreporting and overreporting of browsers in statistics.
What’s left for us is to remember that statistics can be read in different ways – we must be aware of what statistics tell us – and what they don’t tell us. Anything else would be a mistake.