[ericd]

I mean, the numbers are regularly 3-10x off. With that kind of variance, maybe you can pull out some truth if you're the Compete whisperer, but to most people, it's worse than useless, because it can fool the unaware into thinking it means something. If it were at least mostly directionally accurate, that would be a different story, but as far as I can tell, it's mostly a random walk. Alexa is a bit better as far as I can tell, in that it at least conveys somewhat accurate vague trends, but it's still not very helpful. The only one that's worth a damn is Quantcast direct measurement, but then of course that's accurate, and it's opt-in. Comscore's methodology seems better than most, but I haven't compared with hard numbers.

It's especially annoying when people whose opinion of your site matters quotes Compete numbers and trends for your site to you as though they mean something.

There was a really good post that I can't find where Spolsky (I think) was writing incredulously about how Compete was telling him that Stack Overflow traffic had fallen over the past 6 months or so, when it was the opposite, to an extreme.

Seriously, worse than useless.