Hacker News new | ask | show | jobs
by hyperbovine 5823 days ago
It's weird, but as a stats guy, this makes total sense to me. They are sampling a small (hopefully random) group of internet users in order to draw conclusions about how the population at large behaves. It's no different than a political poll. Just like political pollsters, they express their results in relative terms to sidestep the much trickier issue of guessing N, the population size. Hence pollsters do not try to estimate voter turnout, and Alexa does not try to estimate how many people are online.

FWIW, the firms who are giving absolute viewer totals are (I'm pretty sure) just taking this a step further by estimating the total number of people online and multiplying. This is a way harder estimation problem, which explains why the resulting numbers are all over the map. The Alexa data is not any more worthless, just less ambitious.
2 comments
pavs 5822 days ago
Its different from political poll because it is not random. Its like looking at internet usage pattern of Indian internet users and predicting the traffic of Reddit, when most of the traffic of reddit comes from USA/EU/Canada.

And that's exactly whats happening, Alexa relies on people who install Alexa toolbar and from what I remember reading most of Alexa users are from South-east asia, or more precisely, from India.

This is so dubious, its not even funny.

link
penblue 5822 days ago
They have rankings by nation in addition to the overalls. Their base is far from a scientific sample, but I think they give better ballpark figures than some give them credit for.

Their "what's hot" section often has stuff about wordpress and the like, so there are obviously many webmasters. There is plenty of American Idol and all that, also. I have no idea, but I'd guess they advertise the toolbar in different places and maybe allow it to be bundled with screensavers or whatever.

link
cookiecaper 5822 days ago
I think that the only tech-savvy people that use Alexa toolbar are webmasters who are trying to get their Alexa numbers up, basically.
link
gtt 5821 days ago
IOW Selection Bias is the problem here.
link
nirav 5822 days ago
Not so much into stats but isn't a good statistical sampling has to have a uniform distribution over population space like strata/demographics?

I wonder what their sample size is for 1.56 billion internet users (http://www.wolframalpha.com/input/?i=total+internet+users).

As a side note: Alexa is an Amazon.com company!, when did that happen?

link
bbatsell 5822 days ago
1999. If I recall correctly, it was one of their first acquisitions. It didn't make sense to me then, either.
link