Rand Fiskin has created an excellent study of 25 websites, comparing the relative accuracy of external metrics sites (such as Alexa and Technorati) against the actual data provided by analytics programs.
The conclusion that was reached is that these sites provide very little accuracy when compared to actual visitor numbers. However this study gives hard facts with which to back up conclusions that were, for the most part, previously based on anecdotal evidence.
Thanks to Andy Beal over at Marketing Pilgrim for pointing it out.
The study focusses on 25 websites from the niche webdev/tech space including SEOmoz, Matt Cutts of Google and Danny Sullivans Daggle. Kudos to Rand for getting so many sites to give up their visitor data (something which notoriously tends to be kept secret).
The information that Rand presents shows how badly external sites can be skewed depending on their method of obtaining their traffic data. Alexa is an obvious one to pick on as it has always been dogged by allegations of unrealistically inflating tech sites due to the high number of techies installing the toolbar which collects most of their information.
However, it goes on to show that there doesn't appear to be single external service available at the moment that is able to accurately predict which sites are the most visited.
So just remember when putting together a competitor analysis, that although you might like to quote the fact that your site is higher ranked according to whichever provider you prefer, this information really should be considered suspect at best.
http://www.scribes-write.co.uk/article/20061020123734968.html