I don't know why but I saw almost in each famous information site ever using only Net Applications to evaluate the market share of browser. Does it have some sort of better precision rate over the others? It seems strange
Consider three people: one uses browser X and visits Wikipedia twice a day. One uses browser Y and visits three times a day. One uses browser Z and visits once a day. If you count share amongst users, the three shares are equal. If you count share amongst visits to Wikipedia, you get 50% Y, 33% X, 17% Z.
This same problem comes up if you count visits over a variety of sites too. Some biases cancel out (e.g. the sort you'd get if you measured just Huffington Post vs just Fox News), but biases that have to do with some people just browsing _more_ than others in an absolute sense are not.
Of course both numbers, visits and users, are useful information. They just mean different things.