« Are Facebook apps a Long Tail market? | Main | Valleywag's wrong: games want to be free, too »

October 08, 2007



Pretty obvious conclusion... Facebook apps aren't like desktop software. The most successful are the ones that are highly social. And like all social websites, they are subject to strong network effects. The ones many of your friends have already installed are much more interesting, so installs skew heavily toward the most popular apps.

Nick Carr

Robert makes a good point, but I was wondering whether, as well, there's a time lag involved before a large distribution settles into a power law pattern. Facebook Applications are very new, and a substantial portion of them are (I would guess) very, very new. It would seem to me that it would take a certain amount of time for the discovery process to play out, even with good search and discovery tools, and in the early stages the distribution might be more skewed to the head, particularly if a large proportion of the available items are very new. Chris, Have you seen any evidence of such a time lag in the development of the power law pattern?


This sounds too early to tell: as long as I will get Vampire invitation (should ban it, I know), the whole thing will not be mature. To argue with Nick ——hmmr: Mister Carr—— I'd notice that the not-so long-tail considers less then a hundred users (log n = 2): a long tail level, but far from statistical significance. How many apps are they before that drop? A hundred, a thousand? In three month's time? That is amazing! That is code-night worthy.


May be a little below the sophistication of the room here - and I mean that very literally, as I just completed listening to the audiobook of The Long Tail last week - but it occurs to me that the long tail may not show itself here in traffic to apps, but in some other way related to them.

Perhaps we're not quite seeing the long tail of Facebook Apps, but the long tail of Facebook App producers, of app creation creativity, or perhaps even of Facebook page popularity as indexed or expressed by traffic and utilization in relation to the apps. Just a rough-hewn thought, knowing little about the phenomenon myself at this point.

And Chris: outstanding book. Positively outstanding. The audio would have been better had you read it (I almost always feel that way about audiobooks), but the content was great. To actually describe the long tail phenomenon as it affects our world in so many ways is a very important contribution to not only entrepreneurs like myself, but to thoughtful decision-makers in virtually every walk of life.




Nick's comment sounds about right. It's probably too early to say. The facebook platform is only a few months old and the discovery mechanism (the app directory) isn't used much (even by app developers).
I've recoiled from facebook as a user but I'm intrigued by it as a developer. The API makes it possible for facebook apps to be obnoxious about acquiring new users, prompting users to invite all their friends - some applications make this the first order of business.


It looks like if you cut off the data at the top 1000 apps, you probably do have a powerlaw. The drop at the end could just be edge effects because there are only a few thousand facebook apps at this point. If you only take the first 1000 data points, your R-squared fit should go way up.

Kevin Kelly


You have a vocabulary problem. There is obviouslly a long tail of Facebook apps. You can see the "long tail" in the first chart of Tim's you posted. There are lots and lots of little sales.

I'm confused by what you are saying. Are you saying that this long tail is not the "new" long tail you were announcing in your book -- that the sum of the long tail equals the sum of the head? That it is the "old" long tail we always thought of -- that the tail does not really matter?

Do you want to call this old one the long long tail? Or the thin long tail?

And if this is a thin long tail vs a deep long tail, does this mean that old media strategies are best for it?

Chris Anderson


You're right that I wasn't clear enough in this post about definitions. In the previous post I defined a Long Tail distribution as a powerlaw, and discussed the head/tail size implications of that. Powerlaws are, by definition, "fat-tailed", which is why I was suspicious that the Facebook apps were indeed a powerlaw/LongTail as Tim had originally said.

Is that clearer?


Aditya Sood

My supposition about why Facebook apps might not be a long tail may
have to do with Metcalf's Law. Metcalf's Law states that the value
of a network is worth the square of the number of users.

Given that most Facebook apps are network dependent, a user should be
geometrically more inclined to use an app that has more users already
using it. Because your decision making relies heavily on the
decision making of other people, this is fundamentally different
than, say, buying back catalog songs on iTunes which is a single
person's decision.

The mathematics of testing this have sadly receded into my high school
past, but I wonder what the graph of the square root of the users of
those apps would look like, and I wouldn't be surprised if it looked
a little more long tail-ish.


"Power law or not?" vs "Long-Tail or not?" are separate questions. If I understand Chris' thesis, Long Tail is the idea that there is a significant population in the "not-hit" part of the distribution, usually of low volume in any rank, but continuing out to very high ranks.

The idea that there is a region where the distribution is essentially "scale free" seems like the key concept. If we start there, interesting questions include: Can we characterize this region with a power law? And (my favorite), what are the dynamics of the system where scale matters? This last question is at the core of the economics of the long tail businesses in general. For example, determining how inexpensive we make "find" and "acquire" activities corresponds with the "knee" in the distribution.

Scale-free is a misleading mathematical idea in that nothing in nature is actually scale free for all domains. For example, an absurdity of assuming scale-free in every domain WRT music or movie hits is that anything created has at least 1 fan (i.e. we don't have an arbitrarily small hit)--this introduces scale and consequently, the region of arbitrarily small hits with less than 1 fan can't be modeled by a power law. That's a toy case, but illustrates how much scale matters.

Instead of including all the points in the power law fit, maybe we can look at the points up to the knee in power law model (it looks like x~3) and then try to understand what interesting dynamics shape the knee for x>3 with the assumption that some scaling has been introduced by cost, potential audience size limits, or whatever...


all social websites are very helpful,
great script

Honor Gunday

I'd be interested to see what kind of graph groups inside social networks or double-looped viral social networks such as ning have...

It seems like it could depend on the user's behavior.. If they follow the leading groups, it becomes non-longtail, if they individualize they do... am I right?

Do you think you can analyze that too?



It depends on how you define long tail.

I do not feel the existence of a cliff (a curve following a straight line and then falling away) in a log-log graph means that a market is short-tailed. Merely that it is not infinitely long.

It indicates that a scarcity effect or a latent demand exists. But both short-tail and long-tail markets can have latent demand. This could be the 50,000th most popular song in Walmart, the 5 millionth most popular song on iTunes, or articles number 1050 to 250,000 by Jakob Nielsen.

The existance of a sudden cliff indicates that "good findability" exists - people can find most (ie more than 10% - think logarithmically) of your content, and want more.

Even wikis have latent demand. This graph comparing page hits versus rank in several wikis indicate that the curves hit a cliff when the pages are getting between 10 and 100 hits (ever, not per day). (The wikis that hit the cliffs last are anthologia, a one-person show with more effort than popular success, and WikiZnanie, where most of their content is copied by computers from Russian public domain documents)

Getting back to the graph, the main deviation from the power law occurs in the last tenth of an order of magnitude out of 3 orders of magnitude on the x axis. In order to match the line perfectly, you'd need roughly a million facebook applications, 90% of which would be used by fewer than 10 people each. Does the world need a million facebook applications? Do you need to go all the way to the x axis to be long-tailed?

Robert implies that facebook software is more skewed than desktop software. Desktop software has its own network effects, such as new users using software others know how to use, and IT departments preferring to support only product. I'd argue that, at least with spreadsheet software, one software company has by far the majority of active users, a situation not seen with facebook applications.

So compared to desktop software, facebook software is long-tailed.


Devabhaktuni Srikrishna

Here is an updated plot of all Facebook apps with one or more daily active users, not just apps with more than 1000 users which was my earlier plot. The tail seems to have expanded from October 2007. Updated blog post is here.



jeux video

I read both of blogs the previous one and this one too & after read all of this i appreciate that you are true.. Actually the facebook is a social website so most of the people are preferring facebook for website marketing..

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!