Sex, statistics and videotape

I’ve perhaps already “beaten the subject to death”:http://virtuelvis.com/archives/2003/02/properly-measuring-msie – even “more than once”:http://virtuelvis.com/archives/2003/04/msie-not-msie-revisited. Yes; browser statistics.


I have now applied the technique described in those two postings to this site as well, and I am now able to provide you with numbers for by how much overreported MSIE is in my access log for the last week.:
* 2003-07-02: 6.67%
* 2003-07-03: 22.22%
* 2003-07-04: 15.38%
* 2003-07-05: 16.22%
* 2003-07-06: 7.32%
* 2003-07-07: 18.57%
* 2003-07-08: 14.44%
* Average value: 14.40%
The percentage is the percentage of browsers that identified themselves as MSIE, and fetched the stylesheet located in /styles/screen.css without fetching the corresponding MSIE-specific /styles/msie-screen.css stylesheet.
Another number for which I’m only offering the average value: 28% of the visitors that are using Opera, has set their browser to identify as MSIE. I’m actually a bit surprised that the number is that low, given that the default setup in Opera is to fake the UA string.

Previous Post

3 Comments

  1. I appreciate that you publish these measures, Arve, because it proves that most statistics is unreliable. Many websites are using “Redsheriff WebMeasure”:http://www.redsheriff.com/ which basically fetches statistics through JavaScript and as an alternative; images. No serverside checking, no backup for textual user agents, nothing.
    I don’t know how the statistics Redsheriff delivers are being used in each website-company where WebMeasure is deployed, but I know that WebMeasure gives browser/operating system figures which — needless to explain — are as trustworthy as George Bush’s declared reasons for attacking Iraq.
    Since WebMeasure relies upon script and images, it can’t deliver figures from any other user agents than those who supports this, because it won’t get the figures from other user agents at all. And this is of course not good enough.
    I just hope that companies that have deployed WebMeasure don’t use the browser/OS figures for anything other than fun. They are completely unreliable, but if they are used, it explains why so many companies only focuses on Internet Explorer only, as IE of course will have a lot larger pool of users than e.g. Opera (where images and scripts very easilly can be turned off, which isn’t the case with IE).
    The most scary part is that Redsheriff, which most likely is the biggest player in the web statistics area, probably doesn’t know about this issue, and even if they did, don’t tell their customers about it. They don’t know, and they don’t care.

  2. Clever usage of the conditional comments feature.
    One remark: As conditional comments are supported as of IE5 and you are using the downlevel (non-IE and below IE5) hidden version you are excluding IE4 usees from th IE browser groep.
    If you (readers) are interested you can even make a clear divisition in IE versions when you do as follows:
    Make three differently named css files (might be the same file as content).
    Make three conditional comments.
    only IE 6
    only IE 55’s
    only IE 5’s
    (IE4 can’t parse the conditional comments)
    (IE4, I guess one would just use the ueserAgent of the JScript minor version, etc. I don’t think that many Opera browser Identify themselfs as IE4)
    Cybarber

  3. The various versions of Opera has the ability to identify themselves as either IE5 or IE6.
    As for keeping a check on which IE browsers are used, I am keeping track of that. The overwhelming majority are on IE6, outnumbering the IE5.5 users by a factor of ten.
    IE4? I have more visitors using Lynx than IE4. 🙂