Data from Net Applications and StatCounter are different from each other. And it’s not just a little. What complicates browser measurement? Why do the individual measurements of web browser shares differ so much from each other? To what extent can the presented data be trusted? What to watch out for? How do I look at the data that I use to decide which browsers to target for the page I'm about to make? Let’s talk about what skews the measurement so that we can recognize data that doesn't answer our questions.
Different methodologies
There is no uniform way in which data is collected, processed and evaluated. This is subject to what we want to measure. We want to know the share of browsers. Fine, but from where? The proportion of all requests to the servers, or all users who use the Internet? How to distinguish a bot from a user? It’s not negligible, because about 30% of traffic is made up of bots that index content, check the availability of the site, or do anything else. And when it comes to server requirements, it’s probably clear to everyone that some people browse more, some less. The user is identified via a cookie. However, the success of the permanent setup is a chapter in itself. Is the ratio of pages to users the same in all countries? Probably not. That’s why one site in China doesn't represent 20% of the world. So normalization makes sense. However, whether it is reasonable or not depends on what we actually want to know.
Compatibility Mode
Whatever share Internet Explorer has, that fact is not the point of this article. However, it is a fact that its latest version has a minority share compared to its older versions. Yes, you can't install IE 9 on Windows XP, but you also can't install IE 8 on Windows 8. Where am I going with this? This is what the UA string of Internet Explorer looks like in compatible mode. Not the server, but JavaScript, but some tracking codes collect data only from it. Except for cases where the combination of browser version and operating system form obvious nonsense, it is not possible to tell what part of IE 8’s share is made up of a newer version switched to a compatible mode. It is therefore not surprising that it has such a high share in some IE 8 measurements.
Prerendering
Chrome came up with the idea of loading some pages before the user clicks on them. How does he know in advance? He doesn't know and he can't. Some loaded and rendered pages will never be displayed. This is not negligible, it is more than 10% of the pages loaded. Such should not be included in the statistics. For example, NetApplications does this, StatCounter does not. That’s why Chrome has a strikingly high share in StatCounter’s statistics.
Phone or PC
Some measurements apply only to desktop browsers. IE has a high share there, but it must be admitted that desktop data does not correspond to reality. While IE dominates the desktop, Safari dominates the mobile worldwide, but Android in the Czech Republic. The question remains what the shares are between phones, tablets and desktops. So far, the desktop still has 90%.
Browsers on iOS
The assumption that Chrome and Opera have Blink does not apply to iOS. Apple will not allow a browser into the AppStore that is not built on its WebKit core. Statistics of rendering cores rather than browsers would be more useful. But it was created historically, because in the past, each browser had its own core. And so when we look at Chrome’s share on phones, we need to take away the part that Chrome on iOS consists of, because there it behaves the same as Safari.
Tracking Protection and AdBlock
Internet Explorer has a Tracking Protection feature, other browsers have AdBlock. While AdBlock is fine because most filters don't block traffic measurement, having automatic tracking protection enabled in IE blocks any embedded foreign content that appears on multiple different sites. And it doesn't matter if it’s advertising, traffic measurement, jQuery, font, Disqus or Facebook like. The important thing is that it is content from another domain that can set a cookie. It is probably unlikely that the number of people with automatic protection from surveillance is statistically significant. Enabling this feature requires 6 clicks, and you won't find it in the options in the Security tab, but in Add-ons Management, even though it’s a native part of IE. On the other hand, and consistently speaking, if we're not sure how many people have turned on this protection, we can't be sure about IE’s share in Google Analytics.
Incidentally
A number without a unit only makes sense if it is a dimensionless quantity. That is, typically some proportion that is in the range of <0.1>, so it is expressed in hundredths and denoted by % (from the Italian per cento). If someone makes a statement that concerns a certain percentage of (at best, the interviewees) people, it has no informative value if we know nothing about what the whole is.
Finally
Whatever the global statistics are, they may only be important when you're making a new website. If you're iterating, the data measured on your website is important because you know what your visitors are like and who to target. I was surprised that this blog is read by more users in Chrome than in IE (or a lot of them use tracking protection, I have no idea). I'm very happy about that. I've picked up a lot of things here that used to work only in IE, especially on the main page. It led me out of the mistake that I don't have to deal with interoperability so much on this blog. The opposite is true. Interoperability matters.