Figures about web traffic are notoriously hard to interpret - as in, to what extent should you trust them? It can be like trying to discern which opinion poll to believe.
So bearing in mind our story from last night about Facebook passing Google for traffic in the US in a week in March, according to Hitwise, see how you like these numbers: according to Compete.com, in the US in February Twitter's web traffic fell by 9.63%, Facebook's by 4.32%, LinkedIn's by 8.30%, while MySpace lost 11.5% of its unique visitors.
Update: with splendid economy, nutsch points out in the comments that "There were 9.7% fewer days in February compared with January." On which basis Twitter stood still, Facebook actually grew, LinkedIn grew very slightly, and MySpace.. oh dear.
The numbers are posted at Twittercism, and say that Twitter had 21.3m unique visitors in February, and a total of 143.9m overall. Both those numbers are down.
What might you conclude? That Twitter is sooo over? Perhaps - though as the post itself notes, such measures as Compete uses don't include mobile clients or API-based connections. And Twitter's API traffic is the majority of its real traffic - four-fifths of it, according to a presentation by Twitter itself last year. (Which at the time said that if you had 127 followers, you were above average. The average is probably lower now, given that there haven't been many new celebrities joining but lots of normal folk.)
The real problem with the Compete numbers though is that like many web metrics systems they rely on polling people who have toolbar extensions installed, which means that you can't be sure that it's really what happened - only that it represents what happened to the people you polled. Certainly, if you do it well - like polling companies try to do with voting intentions - then it's reliable enough. But just as election polls can be confounded by people who don't respond (or aren't on a landline, the usual system for polling), so toolbar-based metrics can be confounded by browsers that don't take toolbar - notably, mobile-based ones.
Thus is it with Compete, which explains:
"We have a diverse sample of 2,000,000+ U.S. Internet users that have given us permission to analyze the web pages they visit and ask them questions via surveys. We're betting that the insights we create from consumers' online behavior - whether they're watching, searching, shopping or socializing - is valuable for companies who are looking to radically improve their marketing."
Now, 2m internet users is a good slice - so can we be confident that Compete is always right?
The figure showing Facebook falling off is, frankly, strange - although it may be an artefact of the shortness of February plus the bad weather in the US this winter which meant some people, um, couldn't get to work to update their Facebook page to say "I'm at work, geez". (Note: the picture at the top was taken in December, not Febuary.)
The fall in MySpace is in line with expectations. Is anything ever going to pull it out of the spiral towards the internet black hole that is consuming Friends Reunited, Orkut, and the rest?
