Caveats

I see a number of possible problems with these findings. I'm sure readers will alert me to others.

• Opera can change its user-agent string to make look like it's actually another browser (this is useful for fooling some commercial sites that only allow access from certain browsers). I rarely use this feature when I'm browsing with Opera, but it may be popular with other users. For this reason the Opera pie slice might be artificially small.
• Firefox and Mozilla can do the same thing, with an extension called User Agent Switcher. Again, I don't know how often this extension is used, so I don't know what effect this has on the Mozilla/Firefox numbers.
• Opera can turn off referral logging. This breaks certain sites, but I'm sure some percentage of Opera users do use this option because of privacy concerns. Anyone who has this feature turned on will not register in my analysis, which is another reason the Opera figure may be artificially low.
• I'm assuming that user-agent strings that list "Gecko" without any other qualifiers (like "Firefox" or "Netscape") are really Mozilla, but this may not always be the case. (I'd appreciate hearing from readers more knowledgeable about this than I am.) There's a chance that this may have inflated Mozilla's ranking.
• As far as I can tell, AOL 9 is based on the Internet Explorer core rather than Gecko (which is odd, since AOL now owns all of Netscape's intellectual property, including Gecko). But if I'm wrong about this, the Microsoft browser engine figure will be slightly inflated and the Gecko browser engine figure will be a little low.
• Because some user-agent strings include the term "Windows 2000," I'm assuming that "Windows NT 5.x" in the user-agent string signifies Windows XP. But my Windows 2000 number feels low, which leads me to think that some of these apparent WinXP hits are really being generated by Win2K (I don't have access to any Win2K boxes, so I can't test this hypothesis). For this reason, the "OS Family" table (which lumps all versions of Windows together) may be more accurate than the more granular "OS" table.

In the interest of full disclosure, I should mention that I do 95% of my web surfing with Opera, as I have for several years. I'm not wedded to it; if something better comes along I'll happily defect. But I'm sticking with Norway's finest for now because of five features:

• It seems to load pages a smidge faster than anything else. No, I haven't timed it.
• Its "wand" function for storing and autofilling passwords is fantastically useful, and more flexible than the password functions of its rivals.
• The "g" key cycles through showing all images, showing cached images, and showing no images. I've been told by a Firefox developer that this behavior is impossible in any Gecko-derived browser for esoteric architectural reasons.
• The "z" key ultra-conveniently backs up one page.
• It's trivial to turn off plug-ins when you want to save CPU cycles or eliminate hyper-irritating, retina-melting Flash ads. This is a lifesaver on my 1999-era 300MHz Celeron, where every cycle counts.

Conclusion

Remember: this is a quick and dirty analysis of visits to a single low-volume site, so please take these figures with a big grain of salt. There are more detailed, more scientifically rigorous analyses available elsewhere, which I recommend to anyone interested in this topic. Having said that, what conclusions can we draw from my numbers?

The data conform more or less to what I expected, with a few small exceptions. I'm a little surprised that more techies haven't abandoned Internet Explorer, both because of its security problems and because of the added features available in other, more modern browsers. I'm also surprised at the high percentage of hits coming from Windows machines. I would guess that many of these hits come during working hours, when users are browsing with company-issued Windows boxes. An analysis of hits between 8pm and 8am in the browser's time zone might reveal very different distributions for both browsers and operating systems. Finally, I'm amazed at the number of bots out there, scanning and scraping information for who-knows-what purposes.

We can certainly conclude that it would be prudent for web developers to make sure their sites work as expected with browsers other than Internet Explorer, especially if they expect a technically sophisticated audience. Based on the "By Browser Engine" data, it seems clear that developers who test only with IE could be losing up to half of their potential visitors (including anyone with a Linux/Unix/Darwin box)! This is of course an obvious point that competent web designers have been aware of for years, but it bears repeating given the number of sites that still don't look quite right or function correctly outside of IE.

About the author
Chris Cowell-Shah is a Palo Alto-based consultant with Accenture Technology Labs, the research and development office of Accenture. His website is www.cowell-shah.com. He welcomes feedback on this article.