Comscore's audience statistics, which advertising agencies use to decide how to allocate campaign budgets, are already uncomfortable reading for many websites. Compared with a site's own internal logs, the New York company's user panel can appear to significantly undercount monthly visitors. The news just got worse. According to an internal study, the company is double-counting the geekier users who delete the cookies which track their web usage, and thereby show up, repeatedly, as new users. The bottom line? Even Comscore's modest numbers may be overstating a site's audience by a factor of 2.5, on average. And the geekier sites, already the ones that complain most that old-school measurement services miss their gigantic audiences, have the most inflated numbers. Update: I got this one really really wrong. The measurement service was not casting doubt on its own numbers. The Comscore study actually explains why measurement tools that rely on cookies — that is, most internal stats — overcount so much. It's still bad news for geeky sites, in particular, which feel most strongly they're being cheated by Comscore: now there's data to back up Comscore's contention, that the "unique visitors" counted by sites themselves are often far from unique.

Contact information for this author is not available.