Tuesday, March 8th 2005

Browser Recognition of Statistics Services

Over the last few days, having had nothing better to do, I’ve investigated how well the main (free) website statistics services cope with browser recognition. The results of my testing are presented in this table.

The first step was to create a webpage containing all the counters to be studied. I started off with Nedstat Basic 3.0, Statcounter, OneStat Basic 3.0, eXTReMe Tracker, CQ Counter, S-Tracking, Site Meter, NextGenStats, WebCounter and PowerPhlogger. However, the last three turned out to be excruciatingly slow in updating the stats reports and I didn’t have the patience to sit and wait for them. They were excluded from the study, and I don’t feel it was a great loss since the quality in general of those services felt rather low. (WebCounter, for instance, was constantly trying to get you to pay for stuff even though this was supposedly a totally FREE sevice.) After publishing some preliminary results in the user forum at statcounter.com I was asked by David Smith to check out Site Meter too. (I later realized he is the creator of Site Meter.)

The results were collected by loading the counter page in one browser at a time and subsequently checking the browser reported by each stats service. Some services (NedStat, OneStat and CQ Counter) don’t show detailed information about each pageload, which means that I had to track the changes on the browser stats pages available.

For pageloads with non-Windows browsers (Safari, Camino and Konqueror) I received help from Jeff Pony, David Smith and Johann Petrak.

Interesting Findings

There weren’t any really big surprises. However, the myth that “many Opera users are detected as IE users” is at least killed, once and for all. (Opera’s default setting is to identify as IE, so as to prevent the user from unecessarily being locked-out by clueless webmasters.) The only statistics provider to be fooled by the Explorer spoofing is S-Tracking, who seldom (if ever?) drop browser figures on the world OneStat&WebSideStorystyle.

One noteworthy fact is that OneStat’s free stats service doesn’t discriminate between Firefox and the Mozilla Suite (or any other Moz browser for that matter). Yet they publish press releases which do (or look like they do). There are three possibilities here: Either (1) they exclude statistics from the free accounts in their press releases, (2) the browser stats in their press releases are seriously flawed, or (3) they actually count Firefox and Mozilla Suite seperately for the free counters too, but to make a paid upgrade seem worth-wile they don’t report them separately.

Netscape 8 beta had an interesting habit of importing the Firefox user agent string when importing other settings from said browser. The original string includes the information Firefox/0.9.6, which was exchanged for Firefox/1.0.1 after importing Firefox settings. As well as fixing this bug, Netscape need to produce a unique user agent string for the final release if they want to be visible in website stats at all.

Another slightly odd thing was that Site Meter sometimes claimed to be able to see more information in the user agent string than was actually there. One example was Netscape 8 beta in Internet Explorer rendering mode: the user agent string presented when calling navigator.userAgent using JavaScript was Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1), while Site Meter claimed it was Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1) NS8/0.9.6. I can only presume I am missing something here, so if you know – please contact me.

Conclusions

Statcounter is quite clearly king of browser recognition – it accurately reports even the minor versions of almost all browsers. The only thing holding it back from total perfection is that it doesn’t understand Mozilla Suite version numbers. Judging from this test and previous experience it seems like all Mozilla Suite 1.x versions are reported as version 5.0. (The user agent string of Mozilla Suite always begins with Mozilla/5.0.)

Worst of the pack is S-Tracking. For starters it’s the only service fooled by Opera’s user agent spoofing. It also puts all Mozilla-related browsers as well as Safari (!) into one category and simply calls them Netscape. However – because of this – it’s the only service that correctly identifies Netscape 8 beta. Impressive! 🙂

OneStat, which is a fairly popular website tracker, can probably be considered second last. Even if it seldom actually gets it wrong, it’s a pretty blunt tool. It groups all browsers into Microsoft Internet Explorer, Mozilla, Netscape Navigator (including Konqueror!), Opera and Apple (better known to us earthlings as Safari).

No Responses to this post:

  1. Anonymous says:

    Interesting Site

  2. Anonymous says:

    How about checking with some webstat apps that are common? Webalizer, awstats, analog, hitbox comes to mind…

  3. David Naylor says:

    Well, at the moment at least, I don’t have access to a server with logs… Simple as that.

  4. Anonymous says:

    This is very interesting. Are there any others you may be missing, such as Websidestory?I want to work towards getting a more accurate count of browser shares based on the best, most accurate counters, and averaging out.I think this information, research, and your knowledge could help a lot. Please comment on this here, or email me regarding this, bryansmiley78@gmail.com. Kris Silver @SpreadFirefox.com

  5. David Naylor says:

    Websidestory don’t have a free service as far as I know.I’d be happy to help, but I’m not really sure what I could tell you that isn’t in this blog post, or in the linked table. :-)As I wrote above, Statcounter is easily the best and most accurate statistics service when it comes to browser recognition – but they never publish global statistics… 🙁