Nielsen Admits Undercounting Web Traffic
NEW YORK (AdAge.com) -- The Nielsen Co. disclosed today it has been undercounting traffic to websites -- for at least the last three months -- due to a flaw in its system that failed to recognize long internet addresses, underestimating "time spent" on the internet and especially social-media sites.
The flaw affects Nielsen's NetView service, but likely also VideoCensus, MegaView Retail, MegaView Search, AdRelevance, WebRF services as well as custom research. Nielsen believes this flaw has caused the appearance of a 22% decline in time spent on the web over the past several months. The company says the issue will be corrected for December data, which will be delivered in January.
The impact of the error, though, will be felt by the online media and advertising industries, as well as Nielsen. Nielsen data isn't used as buying currency -- like Nielsen's TV ratings are -- but it is used by agencies to plan spending and allocate dollars. Websites or video services that don't have enough scale don't make the plan and don't get dollars. Now, some of them may learn that they've been shortchanged by the biggest brand in media measurement.
In this case, Nielsen's systems were choking on long URLs -- more than 2,000 characters; sessions in which a visitor hit a page containing those long addresses simply weren't counted. These long URLs have proliferated mostly due to social networks, like Facebook, which use URLs to pass data back and forth. However, this flaw would affect websites whether they used long URLs or not: If a Nielsen panelist visited one site with a long URL, all of the other web activity in the session was sometimes lost as well.
"We don't yet know if some properties were more affected than others," said Ari Paparo, exec VP-online products at Nielsen. "That's one of the questions that needs to be resolved."
Media owners have long complained that data from their internal logs was often leagues different than the data they got from Nielsen, and to a lesser extent, Comscore. That discrepancy is usually blamed on the fact that publishers pull audience data from their logs, while Nielsen and Comscore use panels and, more recently, a hybrid panel-census approach.
Nielsen is still investigating the cause of this flaw and extent of the problem and said "every element of our internet measurement methodology, including the panel, collection capabilities and processes" is under review.
"We need to do a better job keeping pace with the rapid evolution of the internet," said the letter, signed by media services president Steve Hasker and Mr. Paparo.
Mr. Paparo joined Nielsen in July after spending many years at Google and, earlier, at DoubleClick. Mr. Hasker, a former partner at McKinsey, joined Nielsen in 2009.
They have also asked the Media Rating Council, which accredits data providers, to review the findings. Going forward, Nielsen will provide progress reports on its investigation, and advises clients to not use the data until the problems are fixed.
Follow Michael Learmonth on Twitter.