Robert Scoble read my post comparing Linden Labs backend metrics for April (and March 07) and Comscore's recent announcement regarding Second Life population and geographical preferences.

"…Almost every entrepreneur I talk to lately whines privately about the stats they see on places like Compete.com, Comscore, and Alexa. Today Tom Conrad of Pandora told me that they are extremely low. He says his service requires registration, so he has very accurate stats of who’s signed into Pandora and he can’t figure out why the stats services are so far off of the real stats.

Marshall Sponder, over on Web Metrics Guru, looks into Comscore’s stats of Second Life’s users and finds the same problem.

The thing is these services rely on toolbars (I can’t even use any of the toolbars on the Macintosh for some reason, and how many of you even have one of these folks’ toolbars loaded? None of my friends do and I’ve been checking). Or they rely on “panels” of Web users that they survey regularly. Do you know the selection mechanisms? How do they know they are getting a representative sample? Clearly very few people who run Web companies find their stats accurate. Yet we’re supposed to believe in them?"

ComScore got caught with their pants down because they reported on Second Life usage where there are backend numbers directly out of Linden Labs that contradict most of the findings.  

When there's no other data than ComScore, Nielsen or HitWise (which does not give absolute numbers, just percentages) to tell us about behavior of other sites, outside our own, or sectors of sites - the inaccuracies are accepted because it's still better than nothing. 

The Linden numbers just show us what we've settled for - an approximation from ComScore that's often way off. Meaning - you don't go to the bank with anything that Comscore, Nielsen or HitWise tells you - you can't - they're too inaccurate.