> Why do so many people get this wrong?

Because it's impossible to get correct. Even Net Market Share can't boast 100% accuracy. The best we can do is form educated estimations.

I think the criticism was that the number from the metric was given the wrong unit (percentage of pageviews vs. percentage of desktop computers), rather than that the metric itself is fuzzy.

Of course such things are educated guesses and have sampling bias, but they aren't even trying to measure the same thing.

This is from an earlier comment: One big difference is that NetMarketShare tries to measure unique clients, while StatCounter measures web page hits.

Since the power users browse (probably a magnitude of order) more web pages than the normal users, Chrome and Firefox is overrepresented in power users who browse a ton.

Toothpaste marketshare analogy. It's possible that 70% of people use Colgate and 30% use Crest, but Colgate sales by volume are only 40% vs. 60% for Crest, since Crest users tend to brush more daily and use more of the toothpaste when they do for some reason.

Or a car analogy: if Toyota sells 40% of cars and Honda only 30%, but 60% of miles driven on roads are by Honda cars since they use it more. Which has a higher marketshare?

"Destop computers"? So how do they measure when someone uses both IE and Chrome on that computer?