Browser statistics are troubled little beasts. They are affected by a multitude of inaccuracies, assumptions and technical limitations. Unless you really know what to look for, you're going to get some dud information.

A lot is being written at the moment about Firefox. Depending on where you look, it's either ripping through the market or it's languishing with just a small band of devotees. Both sides have stats to back their arguments.

i read it on the internet, it must be true

It doesn't take much to figure out that stats are only as good as their source; not to mention that you can make a statistic prove whatever you want:

  • "Satisfaction rates are extremely high, with nine out of ten people recommending this product."
  • "There is a high level of dissatisfaction, with 10% of customers saying they would not recommend this product to friends."

Both statements are absolutely true and backed up by statistics. Hardly elegant pieces of spin, but it's a demonstration of the obvious - it's worth mentioning as a reminder.

Similarly, Coyote Blog reminds us that you should not assume your own personal circumstances are representative of the entire market. Coyote Blog: Browser Market Share? Depends on Who You Ask: I have been a marketer for almost 20 years, and one of the classic mistakes in marketing is to rely too much on your own experience and preferences.

very fast, very dumb machines

On top of human error, there's only so much information that the machines can produce anyway. Computers do not think, they act on absolutely literal information. This is why they can look 'stupid' when they make mistakes a human would never make (but they make them so damn fast!).

When a stats package crunches server logs, it only gets the information that was available to the server. That can be fooled by odd browser identification strings, the vagaries of browser vs. rendering engine and spoofing techniques. Many browsers can identify as something else, in order to get past bad browser sniffing. A few specifics to illustrate the point:

  • Out of the box, Opera is set to 'identify as IE6' in order to get around crappy browser sniffing which only recognises IE and Netscape.
  • Some versions of Safari identify as Mozilla due to a very strange ID string, which gets truncated by most log processors before the 'Safari' bit.
  • There never was a 'Netscape 5', but you'll see it in your logs. Most people attribute it to pre-1.0 versions of Mozilla. I imagine they are correct.
  • Search bots can take up a large amount of your site traffic, throwing the percentages out. I've seen some sites which get more Googlebot hits than anything else, so the percentage of hits from actual humans are lower than they should be.
  • Netscape 8 can fire up a completely different rendering engine - IE or Firefox, for example - and who knows what that is going to do to your logs.

You know, it's a surprise geeks are into coffee and not vodka for morning tea.

throw me a frickin' bone here

Still, everyone gets asked for stats eventually. Developers may tear their hair out at the latest product which only works in IE, but managers will want some proof that IE is not the only browser being used... before you tar and feather anyone. Managers are paid to ask those annoying questions and besides, they have to justify the tar budget later on.

There are a few sources out there:

  • W3Schools | Browser Statistics. W3Schools are quite open that their stats probably show an abnormally low level of IE users, since their target market is web developers.
  • | Global Stats gives statistics via page hit counters. Although serious business sites generally don't have stat counters; this is at least a large sample so it should avoid the sharp differences of sites like W3Schools. It is worth noting - as Stats Weenie observes - TheCounter seems to vary detection rules fairly frequently.
  • Browser News: Statistics gives a rundown of five sources and give a prominent warning about trusting stats. You might leave more confused than you started, but at least that's accurate.
  • Many people cite WebSideStory - Data Spotlight, however I find their level of spin a bit annoying (or maybe it's just that the name sits right on the edge of 'they must be killed' levels of annoying). Currently their two most recent posts are 'Firefox keeps growing!' following their earlier statement 'Firefox gains beginning to slow'.
  • The least biased view probably comes from the Wikipedia: Usage share of web browsers - Wikipedia, the free encyclopedia.

Update: two more sources - Market share for browsers, operating systems and search engines | and W3Counter - Global Web Stats.

the wood for the trees

Despite the warnings of Coyote Blog, you should not ignore your own site logs. If your site has a huge number of Netscape users, then you wouldn't discount it just because TheCounter shows 1% market share. Such anomalies are not unusual in large organisations, where the Standard Operating Environment (SOE) reigns supreme. If two thousand users were given computers with Netscape 6 loading by default, that's probably what most of them are still using.

Yes, I know, it's not very likely that it will be Netscape. Probably an old version of IE. But you get the point. The nice turn-about is when you run a weblog about web development and IE6 is way down the bottom of the list.

so where does this leave us?

Take everything with a grain of salt and you'll be ok. Generally speaking you will be able to find trends in between all the numbers. You just have to watch out for the gotchas and remember that there are probably errors that you can not detect. Ultimately you are better off designing browser-independent sites and only look at stats for amusement.

Finally... I will say from the various stats and logs that I watch, IE has definitely lost ground in the last 12-18 months; with most going to Firefox. I have a site with a very broad audience. A year ago on that site IE had a 95% share, now it has 74%; Firefox has 10%; Safari and Netscape 7 have 5% each; Opera and Mozilla have 2%. Odds, sods and bots round out the figures.

But... well, don't quote me. It's just stats.