« Got Plumber's Butt? | Main | The end of the mass-market record store? »

August 14, 2006



A really great example of a distribution bottleneck, and one that I'm particularly interested in, is the art market. If you're ever going around the galleries up there in SF, ask some of them how many artists they can represent properly. An honest answer will probably be 12 to 20 at best. So with maybe two dozen serious art galleries in the city and thousands of aspiring artists, "shelfspace" becomes a huge issue. It's exponentially worse here in LA, and New York is beyond description.

I'm curious to see how Long Tail economics can apply to this market. Lots of people look at art online, but except for posters, mostly they still go to galleries to buy. Unless artists put their content into purely digital form, in which case I'm not sure if anyone would buy it, it seems like we're still stuck with the bottleneck.

Jason Crawford

Very interesting; I'm glad you posted on this. When you gave your talk at Amazon and showed this slide, my first thought was that the curve looked so regular that it could easily be some different distribution.

I don't personally have the stats mojo to solve this, but if you find the answer someday, please post about it.

Martin Cohen

Techies may know that there is special graph paper that will plot data that is normally distributed as a straight line. A bunch of years ago I wrote plotting software that could do the same for any distribution. It was originally intended for log-normal distributions.

The way it worked was that you gave it a function that computed the density distribution. It then numerically integrated this to get the cumulative distribution and did inverse linear interpolation to make the graph paper and plot the data.

If I still had the source (long gone), I could take data and see how log-normal it is.

The program did NOT try to find the parameters that would make a distribution best fit the data.

Ah, the ghosts of fortran past (on a CDC 6400).

Jakob Nielsen

To visually assess whether a data set follows a normal distribution, plot a QQ chart. (See example of usability data that's indeed very close to a normal distribution, though websites usually have a few outliers.)


One reason for the difference is that distribution isn't the only limiting factor. Imagine a society that has few car dealerships and a media income of $300 a year. Beyond a certain point, no improvement in distribution is going to sell a new car to someone who can't afford it. The same can be said about some item, say a riding power lawn mower, that only a certain number of people can use--those with large lawns.

If you want an example of a rather stupid, artificially created limit on distribution, read what the guy above wrote about the limited number of art gallaries and then take a look at Washington State.

Children's Hospital in Seattle was placing art on its walls and selling it. That seemed a win/win situation for everyone. People got art, while the hospital and artist got much needed money. Poor kids got medical treatment. The state even got to collect sales tax. Then the state property tax assessors stepped in and said that art on the wall (unlike gifts in the gift store or food in the cafeteria) had nothing to do with the non-profit purpose of the hospital. If the hospital wanted to sell art, it had to pay a much higher for-profit property tax for the walls where it sells art. (How do you measure the horizontal square footage of a vertical wall? Is the commercial space just that of the painting, or is it the space of someone standing in front of the painting? How big is that person and should the taxes allow for two people?) Rather than face that sort of messy accounting, the hospital and a number of other non-profits, including my church, quit displaying art for sale.

The result is, of course, a lose/lose situation for everyone. Perhaps it would help if someone would explain the long tail to the twits in our state property tax assessors office.

George Shannon

My master's research in estimate uncertainty suggested that the lognormal BIVARIATE distribution does a good job in modeling the uncertainty associated with the ratio of the actual to the estimate (ACT/EST). The reason I'm posting this is out of curiousity whether or not the results we found for the lognormal bivariate would support use of the lognormal model for market estimates, such as latent value. We performed curve fitting against historical data across a range of product types and level of risk within one organization. Since market valuations are predictions, the lognormal for the ratio of the act/est may be more in line with real-world scenarios (like limited number of screens).

Michael Mitzenmacher

Michael Mitzenmacher here. I'd like to thank you for the compliment of mentioning my survey article. I thought I'd shamelessly self-promote and mention I have an editorial on the future of power law research and the difficult problem of figuring out when you really have a process following a power law in the current issue of Internet Mathematics. The article can also be found on my Web page at https://www.eecs.harvard.edu/ ~michaelm/ListByYear.html in pdf format. (Feel free to look at my other work as well!)

In a less self-promoting vein, there's another very good article on power laws in this issue of Internet Mathematics, and I encourage any academics or researchers out there to encourage your library to subscribe to this new journal.


Wouldn't it make more sense to make your link text meaningful, so instead of linking the word "here":
Jakob Neilsen analyzes the observed fall-off in web traffic on one site here
link the whole phrase:
Jakob Neilsen analyzes the observed fall-off in web traffic on one site
That way automated systems like google can better calculate the popularity of pages.

Barely Technical

Why on earth do you give all generate all your diagrams with Rank on the bottom axis? No one in the real world does that. We all use probability density and (cumulative) distribution graphs. And 'we' doesn't mean hardcore statisticians and the like. 'We' is everyone who has ever seen the infamous bell curve (i.e. normal distribution), and is being forced by you to compare it to a ranked curve.

paper shredders

I am really pleased to post my comment on this blog . It helped me with ocean of knowledge so I really belive you will do much better in the future . Good job web master .

Paper Shredders

I think a lot of organizations are exploring new ways to become more cost efficient these days. Many big companies often use large amounts of paper in their everyday environments.So they need a paper shredders because Paper shredders are a simple solution to destroying many of the documents and materials that hold private information.Taking simple steps, such as purchasing a paper shredder, can help you avoid a lot of pain and suffering in the future. That seems well worth the small investment.Wanted to compliment on your site, it looks really good.

Industrial Paper Shredder

Home shredders are very helpful as well as cheap there are usually old documents and records related to banking and saving sheets.

take surveys online

this is such a good read. Thanks for sharing. I just love your blog i am on here all the time.

pda accessories

This is great. I learned about unconferences a couple years ago, and I got to participate in one recently at WikiSym. I even tried to run one as part of a community event ... but only 24 people came at all, so you can imagine what the unformatted sessions were like.

This is a very important way of communicating. Although it is somewhat demoralizing to list an idea that no one wants to talk about, the small group format is ideal for brainstorming. And the lack of an Expert role in the conversation is so important.

Did you notice any differences with regard to quality of conversation and group size for each topic? It would be an interesting study to see if there is a kind of self-organizing maximum that would surface, or if people would be willing to talk in very large groups because they are more attached to the topic.

The comments to this entry are closed.


The Long Tail by Chris Anderson

Notes and sources for the book

FREE was available in all digital forms--ebook, web book, and audiobook--for free shortly after the hardcover was published on July 7th. The ebook and web book were free for a limited time and limited to certain geographic regions as determined by each national publisher; the unabridged MP3 audiobook (get zip file here) will remain free forever, available in all regions.

Order the hardcover now!