Jul 11 2007
Nielsen/netratings - length matters
A month or so ago Nielsen/Netratings published statistics that seemed to suggest that the average nytimes.com visit lasted an unbelievable 30 minutes. Asking around it seems that the figures really were unbelievable - no-one believed them.
The most significant question for me was “when most people are browsing using tabs is there any way to know how long someone actually spends browsing a web site vs. how long that web site is open in a (forgotten/hidden) browser tab?”
Looking more closely at the netratings the whole protocol Nielsen was using seemed totally suspect. Here is an excerpt from the May report:
Chicago Sun-Times — 1,860 — 0:07:22
Atlanta Journal-Constitution — 1,852 — 0:32:12
So the Chicago Sun-Times was ranked higher than the AJC because their stat for unique visitors was a fraction of a percentage higher. Even though the average visit on the AJC lasted almost 5 times as long!
Recognizing the absurdity of this Nielsen now seems set to switch to “length of visit” as the primary metric.
Have Nielsen discovered some method to determine whether an “open” browser window is actually visible?
The move from “page view” to “length of visit” as primary metric was necessitated in large part by the ubiquity of embedded video. But here again the stats are not as useful as they could be. Are you interested in how many people watch the first 5 seconds of your video, or how many watch the whole thing?
Terry Heaton notes that “there is no single standard for measuring web video“. Flash cue points and Quicktime chapter markers provide simple techniques for ascertaining how much of a video a viewer actually watches. The tubemogul.com study that Heaton references suggests that the larger Video sharing sites are using cue-point methodology to exclude partial views from their counts.
But in order to embed cue-points the video has to be re-encoded using Flash or equivalent. AFAIK few of the simpler flash video encoders support cue-points. More importantly none of the site analysis apps are set up to poll cue-points.
In an environment where length matters we need a few more rulers.
