The atomic unit of attention
I'm a big fan of last.fm, a web application that records what music I listen to and can then find people with similar musical tastes as me. I have set up iTunes on my PC to automatically send data about what music I play, and I get all sorts of interesting information from it.
One of the things last.fm provides is a chart of "Top Artists", based on the tracks you listen to. They even let you put a cool little widget (see left) on your own website showing your chart. I find this pretty interesting, but am getting a bit annoyed that my all-time top artist is "Boards of Canada". I often listen to electronic music when I'm working, I don't like complete silence but don't want something that draws my concentration from whatever it is I'm doing. Boards of Canada are great for this and are a fine band. But I don't think I listen to them any more regularly than, say, Sigur Ros or Mogwai.
The reason for their elevated position in my personal rankings is, of course, the nature of the data that is being analyzed. The album "Geogaddi" by Boards of Canada lasts one hour and contains 22 tracks, whereas "Young Team" by Mogwai also lasts one hour but only has 10 tracks. This means I can listen to Mogwai twice as much, and Boards of Canada would still beat them in my rankings. "Lift your skinny fists.." by Godspeed You!, Black Emperor only has four tracks but goes on for more than an hour. Those guys aren't going to trouble the scorers at all, even though I listen to them a lot.
I suppose a more representative ranking of my musical tastes would be based on the actual time spent listening to an artist, rather than by number of tracks, but I can see that collecting this data would be a much more bandwidth-heavy exercise. It is clearly the easiest data to collect but to measure my attention in discrete tracks isn't giving a true representation of things.
This is also true in other forms of media. I'll often spend longer reading one or two in-depth features on the Guardian website of a lunchtime than it takes me to scan twenty or so stories on the BBC website. Yet, with the atomic unit of web attention being the page view, the BBC get twenty times the number of "hits". And with more video content available on the web, you can't go accurately measuring these things by word count.
Attention analysis is still in its infancy, but clearly needs to start coming up with ways that better represent what people are actually paying attention to. It will be interesting to see what techniques start to appear.
A possible way of sorting out last.fm's rankings is to apply an average tracks/hour weighting to each artist, meaning the same data can be collected in the same efficient manner, but the generated statistics are more meaningful. I imagine we'll be seeing all sorts of techniques borrowed from other areas of statistical analysis. The ICC cricket player rankings are all weighted - a batsman scoring a century against weak bowlers on a flat track contributes less to his rankings than a fifty against world-class bowlers on a fifth-day turner. I'm looking forward to the day when I can categorically prove whether Kevin Pietersen is better than Radiohead!