Weighting for Good Web Stats

I always like to look at numbers for the market share of GNU/Linux. I came upon SJVN’s article about FireFox slipping against IE, in which he passed on the explanation from Net Applications about their weighting of hit counts.

What immediately hit me with great clarity is that the weighting as described is seriously wrong! If you are counting browsers and operating systems, the number of users of the Internet is irrelevant. For example, in many schools there are from 3 to 10 users per PC. In China the ratio of users of the Internet is estimated by the CIA to be 389 million, more than the USA, with 245 million while PC shipments to China are about 18.4 million in Q3 2010 while USA shipped 17.6 million. That is the ratio of users to PCs is much higher in China than the USA. Homes in USA often have multiple PCs while in China many use Internet cafes. The result, Net Applications is counting PCs/browsers multiple times in China. In fact, growth of shipments of PCs in China is so large recently, the number of operational PCs in China is much less than in the USA, 99 million compared to 264 million in 2008.

The same thing is happening everywhere on the globe. Students or employees often use one PC for two or more users. So, weighting by the number of users is wrong.

Now, China has a high usage of GNU/Linux compared to Canada or the USA but, if the client sites of Net Applications are more likely to be visited by businesses or organizations using XP than GNU/Linux, overweighting them could certainly exaggerate the tenacity of that other OS share.

Of course, attempting to find the shares for China leads to “The report you requested requires that you are a subscriber to “Geolocation Upgrade”.”. Information is more fun when it is free.

About Robert Pogson

I am a retired teacher in Canada. I taught in the subject areas where I have worked for almost forty years: maths, physics, chemistry and computers. I love hunting, fishing, picking berries and mushrooms, too.
This entry was posted in technology. Bookmark the permalink.

3 Responses to Weighting for Good Web Stats

  1. Ray says:

    “OEMs of course could tell us precisely how many PCs they shipped with what OS.”

    I wish though, but we also have to include people who built their own PCs. 🙁

  2. Thanks for the link. Wikipedia also has http://en.wikipedia.org/wiki/Usage_share_of_operating_systems which shows wide variations depending on sources of information.

    I find it interesting that folks who build supercomputers and can choose their OS, choose GNU/Linux by a ratio of 90%:1% versus that other OS. I also find it interesting that even Wikimedia which is unbiased with respect to OS but biased in language shows one of the lowest counts for that other OS but almost doubles the published unit shipment share of MacOS. I wish ISPs or Google or someone with global reach would do the analysis right. OEMs of course could tell us precisely how many PCs they shipped with what OS.

  3. Dan Serban says:

    Looking at this:

    In time, I have learned to give less credibility to those statistics that attempt to shed positive light on Microsoft’s POS browser.
    You can see clearly that Net Apps is an outlier (I call it “outliar”) claiming an IE market share in the mid 50s while everyone else shows it to be in the low to mid 40s.
    Of all the browser makers, Microsoft is the only one who attempts to distort statistics in order to distort perceptions in order to preserve their monopoly.
    That is because Microsoft’s only chance at surviving in a market is to have a monopoly on it.

    So the next time you see browser statistics where Net Apps figures have been included in the mix, do yourself a favor and run a spreadsheet without Net Apps to get a picture that is closer to accuracy.

Leave a Reply