Ross Mayfield, Chris Gulker, Doc Searls, and Azeem Azhar have an interesting discussion going on the use of power laws as a model for link and traffic distributions in the Web. This was apparently kicked off by a recent paper from NEC Research, which in turn builds on a thread of research started by Lada Adamic and Bernardo Huberman when at Xerox PARC, now at HP Labs. (Check here for a starting place in this literature.) Some of this has made its way into the popular press under the heading of 'winner take all on the web', since a power law distribution will generally give you a few sites with the lion's share of the inbound links and traffic. One of the NEC paper's interesting contributions it to tease out various commercial markets on the Web, and note that coefficients of the power law differ by market, and some even deviate significantly from the big winner rule.

I've been intrigued with the Power Law view of the Web for some time, and even gave a short talk about it at an Institute for the Future panel with Jim Herriot two years back. So I'll pitch in a few nitpicks about the current study and discussion, and a couple of hopefully more substantial points borrowed from that old talk:

Nit #1: I would generally prefer Adamic's approach to analyzing actual browsing patterns to static link checking, particularly when trying to apply this work to the blogosphere. There are lots of dead dog archive pages and moribund blogs out there. Now, a study that showed the changing browsing distributions over time, that would be cool! Is there 'class mobility' on the Web?

Nit #2: It's also possible to overanalyze the individual and group behaviors and motivations leading to the power law pattern. Somewhere in my attic is moldering an old notebook containing link statistics from one of the grand-daddy hypertexts: it was generated in 1986 from the old Grolier electronic encyclopedia, by text matching the see also links between articles. Care to take a guess what the link distribution looked like? Uh-huh. You can get the same pattern from a few paid editors working in parallel as with the infinite number of monkeysauthors on the Web.

To me the more interesting analysis is to overlay commercial considerations onto the Power Law distribution, particularly when thinking about the Web as a medium or a market.

If you look at power law curve of traffic (attention) distribution, you can ask a question about the commercial viability of each place on the curve, that is, can you make a business out of that position? Your answer will depend on the medium in which you're working, and compares things like fixed costs of infrastructure, distribution access, and variable costs of creation to the revenue you can extract from the audience. In any commercial medium, then, the power law distribution is actually truncated. This makes a big difference.

Consider the effect of the transition from terrestrial broadcast to satellite distribution of television. Once upon a time, there were three broadcast TV networks. Now there are dozens to hundreds of choices. There are still big winners, but the win (share) isn't as big, because the tail of the attention distribution curve has grown.

On to a slightly more controversial topic, the music biz. Fixed costs of infrastructure? Down. Variable costs of creating content? Down. Access to distribution? Thereon hangs a tale, as they say, because it's the last defensive element of an industry that depends on keeping the curve truncated, and the winners big.

Blogosphere? It's just one big vanity press here, so the attention is all we're likely to expect and get. The tail of the curve is very long. You might think it's Glenn Reynolds up there at the head of the curve, but it's not. It's CNN. Because every minute you spend reading this blog (thanks!) or the Instapundit is one that the big commercial boys don't get. I'll come back to that issue below.

Now being bloody-minded by current vocation, I'd like to ask what kind of businesses result at various points on the curve. Particularly, can you get a venture style exit from them, that is, 'do they scale?'. It shouldn't take too much business insight to realize that the vast majority of positions on the curve, if viable at all, are the cyber equivalent of mom-and-pop shops. Cash flow positive, providing a decent living to the operators and some enjoyment to the customers. Their territory is defined by specialized interest rather than by geography. No exit for folks like me. At the top of the curve, you can get an exit, if you were one of the first comers and marketed like hell. This is where the type of accretion discussed in the research comes in. So maybe there was something to the 'first mover advantage' after all, but too few chairs when the music stopped. Game over.

But how about businesses that influence the shape of the distribution, either flatten it out or extend the tail, rather than take a particular position on it? Can you define a business where the value grows as an result of growing the commercially viable part of the distribution? I think so, and I call these 'diversity businesses.' Google is a diversity business, the bigger the Web the more you need it. eBay is certainly one, it's value related to the total size of the cyber garage sale. I will also argue that at least the book portion of Amazon is a diversity business - it's always marketed on selection. (And that's why it's a bit deceptive to list Amazon as a singular 'big winner'). One may even argue that every ISP is in a diversity business, because the value of 'IP tone' grows with the reachable sites and variety of content. The music industry lives in constant fear of a sustainable diversity business erupting in their medium. Creating a diversity business that grows by building a media and market platform can be a scalable proposition. Assuming you can find a way to extract revenue, of course.

And finally, a few words on the pernicious effects of bundling a diversity business model together with a model based on trying to occupy one of the few slots on top of the curve, that is, being a broadcaster. Come to think of it, I already wrote them.
5:53:32 PM