It's market share/installed base/whatever time! Net Applications' figures had been under review for a while, and now we know why: they've finally done what should have been done ages ago. Namely, they've added country-level weighting to their figures. They also added this retrospectively, meaning all previous figures have been updated to reflect the change. This has devastating effects for some.
To read all comments associated with this story, please click here.
Member since:2006-04-28
Well, they probably have statistics from let's say 10.000 English sites. And statistics from 100 Chinese sites. So if for example, in the English sites 10% of the visitors are using OS X and on the Chinese sites, only 0.1% are using OS X, the final statistics will say that ~10% of the world wide population uses OS X. How accurate is that? If China has, say, 300 million internet users, it is unfair to not give more weight to each of those 100 sites in Chinese than to each of the 10.000 sites in English. I don't say the math is perfect, but it's better than it was before for sure.
Member since:2005-07-24
I should have read the linked article, it explains what they did well enough to actually make sense ( the meaning of country-level weighing just didn't click right ).
I should make a point to always look deeper when something that should be, relatively, simple seems to be done so wrong...
--The loon
hrm.. now only if I could remember the correct term for the weighing algorithm... oh well.
Member since:
2005-07-24
Why.. oh why are we normalizing in reference to something unrelated??
If you want a market share value, find the size of the overall market and find out what percentage of that market does something. Simple.
NOW, we decide to weigh people differently? Stupid statisticians are always screwing this up!! If you are looking at GLOBAL trends, you look at GLOBAL values.
ONLY when looking at LOCAL/NATIONAL trends do you break down the data into those locales. Normalizing the data into a mean on false pretenses is just bad math! Oh.. wait... statistics is all about ignoring some known data in order to guess at the real value...
So what data is missing? The actual global number of users. Well, take the CIA data, add it all together, then find your share directly. Duh. No reason to break the data down into all these locales and giving each locale a different weight.
All that accomplishes is counting a single China-man's visit as 2/3 of a visit, and counting an American-man's visit as 1 1/3 of a visit. Kinda dumb.
--The loon