Alexa is a panel based service (around a million I believe, out of 204′ish mm US internet users), meaning they take a small slice of internet users that they gather data on and try to extrapolate those visiting habits out to the internet population as a whole. As you can imagine, this is very difficult to do accurately, especially for smaller sites that don’t have a lot of reach.

Every panel also has inherit biases that must be corrected for. Perhaps Alexa’s panel happened to skew heavily towards tech enthusiasts, so tech sites were being over-represented. It’s actually likely that they simply improved their panel inference algorithm which better corrected for particular biases.

I’m no fan of Alexa and panel based measurement for the Internet, but I think Alexa is smart enough to know that any hint of favoritism towards any site would be its death blow.

But in general I believe we already need a tool that will represent actual influence instead of traffic to use its rankings when talking to advertisers because they often buy reputation for their brands instead of traffic itself.