Ben Lorica

One week^† into its public launch, the Google Apps Marketplace has just under 1,500 (enterprise) apps. Combined with Salesfore.com's app exchange (also with over a thousand apps), enterprises interested in moving to cloud apps have an increasing number of software tools to choose from.

Popular apps (measured in terms of # of installs) includes graphic design and office integration apps (aviary design suite and offisync), a collaboration and project management tool (manymoon), a free travel planner (tripit), a basic ERP app (myerp.com), and a CRM application (Zoho CRM).

The typical supplier has about 2 offerings in the Google Apps Marketplace. Below are the suppliers with the most number of unique apps:

(†) Data for this post was through 2/16/2010.

Asia's share of the more than 400 million active Facebook users recently surged past 15%:

With a market penetration of 1.7% in Asia and Africa, the company has barely scratched the surface in both regions. While the company continued to add users in Southeast Asia, there were an additional 2.3 million users from South Asia over the past 12 weeks. In fact according to Alexa, Facebook has already overtaken Orkut in India. It didn't take long for Facebook to threaten Friendster's leadership position in Southeast Asia so something similar was likely to happen in India. But I thought it would take them longer to overtake Orkut^† in India.

The share of users from the Middle East / North Africa remains stable (at just over 8%) and the region had the second fastest-growth rate over the past 12 weeks:

As was the case in my previous post, the share of users age 18-25 remains higher in regions outside the U.S., especially in Asia, the Middle East / North Africa, Africa, and South America. [For recent growth rates by age group, click HERE.]

While Asia and the Middle East are the fastest-growth regions, Facebook continues to add users everywhere. Eastern Europe continued to be fertile territory, with the company close to doubling its active members in Romania (up 86% over the last 12 weeks). Below is a list of fastest-growth countries in each region:

(†) Speaking of Orkut, for what it's worth, Facebook added 800,000 active users in Brazil over the past 12 weeks.

In an earlier post, I examined the average price of the Top 100 PAID apps and noted that the relationship between price and popularity was somewhat dependent on the category. But in the Book category, I concluded that the Top 10 PAID apps were on average cheaper than those ranked 91-100. But what if we examine all Book apps, will the long tail apps be pricier?

The animated graphic below traces the evolution of prices in the iTunes Book category. In Q3-2009 the Book category exceeded 10,000 PAID apps, and since then long tail Book apps have (on average) tended to be much more expensive than their more popular counterparts^†.

Since there are far fewer FREE apps compared to other large categories, pricing is especially critical for Book apps. There are now over 28,000 Book apps, 92% of which are PAID apps^††. Looking ahead, the iPad will be available in a few months and many publishers will need to learn how to price their apps for yet another device (see for example [1], [2]).

[For more on ebooks and electronic publishing, be sure to follow events at this week's TOC conference on twitter.]

(†) There is an upward trend in MEAN price from the more popular apps to the long tail, indicating that many more pricey book apps are in the long tail. The graphic also nicely shows the evolution of prices over time.
(††) The animated graph ends on 2/14/2010, at which point the chart represents the 26,000 PAID Book apps available in the U.S. iTunes App store during that week. In comparison, there were over 29,000 Game apps, 70% of which were PAID apps.

Last week marked the first time the U.S. iTunes store had over 150,000 apps available. Close to 31,000 different developers (or "sellers") were responsible for those apps, with many offering one to five apps, while a few offered over a hundred different apps.

Which developers consistently produce top-selling apps? I examined the percentage of apps produced by a developer that became best-sellers. To identify best-selling apps, I used the Top 100 Free and Top 100 Paid, and the recently launched Top 100 Grossing apps lists.

I've noted that Games dominate these Top 100 lists, so it's no surprise that Game developers are among the most efficient producers of best-selling PAID apps. A pair of large Game developers (Gameloft and EA) offered over 40 different PAID apps over the last year, yet managed to have 3 out of 4 of their apps land on the Top 100 PAID apps list. The typical large^† developer only had 1 out of 10 apps (9%) appear on the Top 100 list. (NOTE: In each of the graphs below, I only show the 25 most efficient developers. The MEDIAN Efficiency is for all developers that had at least the stated number of apps during the period.)

Game publishers also dominated the list of most efficient FREE app developers, but a pair of (adult-oriented) Entertainment developers were among the most consistent producers of popular FREE apps. Also note that a different set of Game publishers are producing the best-selling FREE games:

Apple launched the Top 100 Grossing apps list in September 2009, so there isn't as much historical data available. There are quite a few individual developers who've produced top-grossing apps. Given that a few small and successful Game developers were acquired last year, small outfits who consistently produce best-sellers are attractive acquisition targets. The graph below is limited to developers with at least 5 PAID apps since September 2009:

[By using popularity rankings within a category, one can also identify the most efficient developers for individual iTunes app store categories.]

As far as embracing the iPad, several developers listed above are very enthusiastic about producing iPad apps. The interest is particularly high among iPhone Game developers^††, I would really be shocked if Games aren't a major component of the iPad app ecosystem.

(†) For this post a large developer (will usually) refer to one that offered over 10 Paid or Free apps from Feb/2009 to Feb 7, 2010.
(††) See [1], [2], [3], [4]. The other interesting thing I noticed from the data is how many of the key iPhone Game publishers are located in the SF Bay Area.

One of the largest gatherings of mathematicians, the joint meetings of the AMS/MAA/SIAM, took place last week in San Francisco. Knowing that there were going to be over 6,000 pure and applied mathematicians at Moscone West, I took some time off from work and attended several sessions. Below are a few (somewhat technical) highlights. (It's the only conference I've attended where the person managing the press room, was also working on some equations in-between helping the media.)

The Machine-Learning Bubble in Computational Medicine (Challenges in Computational Medicine and Biology)
Donald Geman gave a nice survey of the problems and mathematical techniques frequently used in computational biology. He also raised something that struck a chord with me. While computational biology has things in common with other fields ("small n, large d problem": small samples, relative to the number of dimensions), techniques that work in fields like computer vision don't automatically translate to biology. First, the size of samples in biology and medicine are orders of magnitude smaller compared to other fields. Secondly, while black boxes (think SVM's or neural nets) are acceptable in other fields, biologists want accurate predictions and explanations for why/how algorithms work. Finally, it isn't clear if there are underlying low-dimensional structures in biological data. Taken together, Geman wonders if machine-learning's possible role in biology and medicine has been overhyped.

Using Unlabeled Data To Identify Optimal Classifiers (A Geometric Perspective on Learning Theory and Algorithms)
Revisiting, the "small n, large d" problem, Partha Niyogi gave an overview of recent geometric approaches^† to machine-learning. In order to mitigate the curse of dimensionality, Niyogi and his fellow researchers exploit the tendency of (natural) data be be non-uniformly distributed. In particular, they use the shape of the data to determine optimal machine-learning classifiers. In their version of manifold learning, they assume that the space of target functions (e.g. all possible classifiers), consists of functions supported on a submanifold^†† of the original high-dimensional euclidean/feature space. One of the most interesting features of their geometric approach, is their use of both labeled and unlabeled data^††† to identify optimal classifiers. The traditional approaches to training classifiers require labeled data. So while one can use mechanical turks to increase the amount of labeled data for learning purposes, the geometric techniques outlined by Dr. Niyogi actually take advantage of any unlabeled data you may already have. Lest you think that these are purely academic/theoretical techniques, Dr. Niyogi cites a company that uses these algorithms to analyze and classify child speech patterns. With so much Data Exhaust available, I can't help but think that techniques that can leverage unlabeled data will prove useful in many domains. (Niyogi and his collaborators have many papers on Manifold Learning, including one that describes the algorithms, and another that provides the theoretical foundations.)

The Calculus of Friendship
Mathematician Stephen Strogatz is known to many Radar readers for his work in network theory ("small-world networks") with his student Duncan Watts. I went to his talk thinking it would cover recent developments in random graphs. The talk turned out to be about his recent book chronicling his long friendship with his high school math teacher. What started out with letters that talked only about calculus and math problems, evolved into a deeper relationship over the last decade. The letters ranged from humorous calculus problems, to moving personal correspondence. For a preview of his book, listen to this recent Radiolab segment featuring Dr. Strogatz and his teacher:

[What made his teacher into a great instructor/mentor? Dr. Strogatz mentioned a few characteristics, many of which could be be re-purposed into advice for business leaders and managers. Yet another reason to read his book.]

Geomathematics (Mathematics and the Geological Sciences)
Another highlight of the conference was a symposium devoted to the emerging field of geomathematics. Given that the geological sciences routinely deal with Big Data sets, developments in geomathematics are worth paying attention to.

(†) To illustrate how geometric these techniques are, Niyogi outlined versions of the Laplace-Beltrami operator, the Heat Kernel, and Homology in his short talk. I went to another interesting talk on geometric structures and discrete graphs, but from what I could gather, it was mostly theoretical in nature.
(††) Niyogi and his fellow researchers assert that "... for almost any imaginable source of meaningful high-dimensional data, the space of possible configurations occupies only a tiny portion of the total volume available. One therefore suspects that a non-linear low-dimensional manifold may yield a useful approximation to this structure."
(†††) In classification problems, labeled data are ordered pairs of feature vectors and their corresponding class labels. In the geometric approach to learning classifiers, unlabeled data can be used to recover the "intrinsic geometric structure" of marginal probability density functions.

Next week, O'Reilly's Research Director Roger Magoulas, will lead an exciting panel discussion on Big Data^†. The focus will be on the piles of data that companies have been collecting, and are just beginning to analyze:

The internet and social media create a mountain of random, unstructured, and at times ephemeral data by-products, which may appear to be trash. Yet, one person’s trash is another’s treasure. From FaceBook to Netflix, people are spending more time sharing their thoughts, opinions, plans and perspectives as they socialize and conduct business online. With each of these Internet exchanges traces of information,or Data Exhaust, are left behind. When correlated or combined, these snippets can provide insight into political views, professional achievements, purchasing behaviors, and demographic information—pinpointing trend setters and leading indicators. Brilliant innovators now re-purpose this data stream, aggregating and analyzing the data to provide new products or services.

(†) Recent Radar posts on Big Data: (1) Counting Unique Users in Real-time with Streaming Databases, (2) Pipelining and Real-time Analytics with MapReduce Online

Measured in terms of number of unique apps^†, the Top 5 categories in the U.S. app store have been Games, Books, Entertainment, Travel and Utilities. But comparing categories in terms of number of apps doesn't capture the challenge of developing applications in different categories. As I noted in an earlier post, it's much easier to develop a Book app than an interactive game.

One crude measure for the relative complexity of developing apps across categories is to compare the number of apps per seller. The Top 5 categories in Nov/2009, were Books (17 apps per seller), Travel (6 apps per seller), Education (4 per seller), Reference and Sports (3 per seller). There were also 3 apps per seller in the Games and Entertainment categories in Nov/2009:

With Facebook topping 330 million active users over the past week, the company's strongest growth region continues to be Asia. Over the last 12 weeks, Facebook added close to 17M active users in Asia alone. Since my previous post, the share of active users from Asia grew by 2% (to 13.5% of all users), and roughly 1 in 7 users now come from the region. With a market penetration under 2%, Facebook is poised to add many more users in Asia (and Africa).

Compared to the U.S., the proportion of Facebook users in their teens (13-17) or in the 18-25 age group are much higher in Asia:

As was the case in other parts of the world, expect the share of users 45 and older to climb as Facebook becomes more mainstream in Asia. Growth was strong across all age groups in Asia over the last 12 weeks, particularly among teens (+90%) and the 18-25 age group (+60%).

In closing I want to highlight countries (within several regions) where Facebook has been growing rapidly:

In Europe, growth has been fastest in the East: as an example, the number of active users in Poland doubled over the last 12 weeks. Growth in Southeast Asia remains strong in countries that have been home to Friendster's core user base. While Facebook added over 800,000 active users in Brazil, for now Orkut remains the dominant social network in South America's most populous country.

In a previous post I described SQL databases designed to handle data streams. In their latest release, Truviso announced technology that allows companies to track unique users in real-time. Truviso uses the same basic idea I described in my earlier post:

Recognizing that "data is moving until it gets stored", the idea behind many real-time analytic engines is to start applying the same analytic techniques to moving (streams) and static (stored) data.

Once companies can do simple counts and averages in real-time, the next step is to use real-time information for predictive analytics. Truviso has customers using their system for "on-the-fly predictive modeling".

The other main enhancement in this release is Truviso's move towards parallel processing. Their new execution engine processes runs or blocks of data in parallel in multi-core systems or multi-node environments. Using Truviso's parallel execution engine is straightforward on a single multi-core server, but on a multi-node cluster it may require considerable attention to configuration.

[For my previous posts on real-time analytic tools see here and here.]

While it might be true that the number of Book apps is growing at a faster rate, Games continue to dominate the list of popular U.S. iTunes Apps. Games accounted for about a fifth of all iTunes apps over the past week^†, but the category continued to have a disproportionate share of the Top 100 charts, accounting for 52% of the Top Grossing, 56% of the Top Paid, and 50% of the Top Free apps:

Since most Book apps are actually individual e-books, the Gaming category would have a hard time keeping up with the ever increasing number of Books. Once publishers figured out how to turn their titles into iPhone apps, the number of Book apps started growing faster than Games. Nevertheless Games continue to rule the Top 100 charts.

A similar story is playing out on the Android platform: the most popular Android apps are primarily Games. (In the Android taxonomy, most Books are in the Reference category.)

Returning to the top iPhone apps, the price of the Top Grossing apps stabilized somewhat last week. Except for the top decile (rank 1 through 10) for which the median price was about $7, the median price across the other deciles was around $5.

Over the last week, the Top Paid Games were slightly more expensive than apps that made the overall Top 100 Paid list. iPhone Game developers will tell you that (visually) compelling and engaging iPhone Games are far from trivial to design and market^††. So it's no surprise that the creators of the most popular Games are starting to charge a little more for their software.

(†) Data for this post was for the week ending 11/1/2009.
(††) First, designing for such a small screen poses a major challenge. Secondly, the sheer number of Game apps (close to 20K last week) makes it hard to create something that turns into a long-running top-seller.

I took the set of users^† who posted tweets containing the hashtag #w2s and determined who those users followed. Unlike the list of the most followed users in all of Twitter, the list isn't dominated by celebrities. (A few coders landed in the top 50.) Regular Radar readers will be familiar with many of the users listed below: over 20 of the top 50 are based in the SF Bay Area. Of the over 700 users I identified, a third follow Tim:

Some organizations create their own real-time analysis tools, while others turn to specialized solutions^††. The Huffington Post developed in-house tools that let editors optimize headlines in near real-time. In some domains, the need for real-time analytics isn't new and companies have moved in with targeted products: SF-based Splunk is a popular real-time analytic tool for IT organizations.

In a previous post, I highlighted SQL-based real-time analytic tools that can handle large amounts of data. Tools like Truviso (based on the Postgres database) and streambase are attractive in that they require little adjustment for developers already familiar with SQL. In the same post, I noted that other big data management systems such as MPP databases and MapReduce/Hadoop were too batch-oriented (load all the data, then analyze) to deliver analysis in near real-time.

At least for MapReduce/Hadoop systems things may have changed slightly since my last post. A group of researchers from UC Berkeley and Yahoo recently modified MapReduce to allow for pipelining between operators. Rather than waiting for a Map or Reduce operator to complete (or "materialize to stable storage") before kicking off a subsequent operation, their solution is to modify MapReduce to allow intermediate data to be pipelined between operators. As they noted in their paper, pipelining holds several advantages:

A downstream dataflow element can begin consuming data before a producer element has finished execution, which can increase opportunities for parallelism, improve utilization, and reduce response time.

Since reducers begin processing data as soon as it is produced by mappers, they can generate and refine an approximation of their final answer during the course of execution. This technique, known as online aggregation, can reduce the turnaround time for data analysis by several orders of magnitude.

Pipelining widens the domain of problems to which MapReduce can be applied. This allows MapReduce to be applied to domains such as system monitoring and stream processing.

In a recent conversation with lead authors Tyson Condie and Neil Conway, they highlighted a few other features of HOP that would make it attractive to current Hadoop users. First, HOP not only preserves Hadoop's public interfaces, it also allows for jobs to be co-scheduled and pipelined, thus reducing the need to write results to HDFS. Second, pipelining leads to preliminary results and early feedback, resulting in faster debugging cycles. Upon seeing early results, a developer can either kill a task, or toggle between pipeline and block mode. Third, HOP does a better job of handling stragglers (slow running tasks) by using previous results to kick-off smart re-starts. Finally, they are currently incorporating a continuous and adaptive optimizer that for a given task, will let HOP converge to the optimal degree of parallelism. The optimizer will allow HOP to scale up/down, dynamically adding/dropping mappers & reducers, based on data being pipelined. In preliminary experiments, they found that superior cluster utilization via pipelining can mean substantial reductions in job completion times.

For those interested in performing real-time analytics within Hadoop, Tyson and Neil informed us that they will make the HOP code publicly available within a month. When asked if HOP can handle large data sets, they confirmed that researchers inside Yahoo have ongoing (successful) experiments using HOP on "Hadoop scale" data. Over the long-term, they predict some form of pipelining will become standard within Hadoop.

So how does HOP compare with the real-time SQL databases I described in an earlier post? For domains where the latency required is in the order of (sub) milliseconds (e.g. algorithmic trading), HOP probably won't help. OTOH, solutions like Truviso and streambase have shown they can handle those types of problems. But for a broader class of problems where a delay of a few seconds is acceptable, HOP will be a suitable analytic engine. In terms of usability, tools like Truviso and streambase look and work like standard SQL, making them fairly accessible to a broad class of users. To make HOP more accessible, Tyson and Neil noted that one interesting side project is to modify equivalent MapReduce tools (Hive and Pig) to incorporate "continuous and real-time queries".

(†) Traditional pull-oriented sytems require subscribers to nag publishers regularly ("Do you have something new?"). Push models deliver content to clients automatically as soon as new content is published ("Don't call us, we'll call you.").
(††) For real-time structured data analysis, enterprises favor the term complex event-processing (CEP). An example is TIBCO's CEP software.

Mechanical Turk service provider CrowdFlower^† and microwork non-profit Samasource have teamed up to make their services available to iPhone users. Users of CrowdFlower's mechanical turk platform can now opt to send their tasks to iPhone users. Previously, CrowdFlower users could choose between Amazon mechanical turks or CrowdFlower's stable of turks.

The Give Work iPhone app takes tasks (created by real companies) and sends it to iPhone users who volunteer to complete them. Meanwhile, workers in a Kenyan refugee camp perform the same tasks using CrowdFlower's regular web interface. In essence, Kenyan refugees work to increase the accuracy of the results provided by the army of volunteer iPhone mechanical turks. In a previous post on Mechanical Turk Best Practices, I highlighted recent research that suggested that for a large set of tasks, the aggregate work of 4-6 turks compare favorably with a single (domain) expert.

The payment for tasks sent to CrowdFlower's iPhone app goes entirely to the workers in the Kenyan refugee camp. In addition, Samasource has negotiated with money transfer services, so the payment goes through with zero transaction costs.

The turks in the refugee camps are recent graduates of Samasource's computer training program. Rather than sitting idly while they wait to be employed, they earn money performing simple computer tasks for real companies. On the other hand, Give Work app users volunteer to perform simple tasks on their iPhone knowing that refugees in Africa are benefiting. CrowdFlower founder Lukas Biewald notes that their work with Samasource opens up their platform to companies who want to tap into and help micro-workers in developing countries.

There are other mechanical turk services that employ workers in developing countries (see for example txteagle). What distinguishes CrowdFlower is an innovative web interface that lets companies easily upload/define their projects and choose the set of turks they want to use: Amazon, CrowdFlower, and now iPhone users + Kenyan refugees. CrowdFlower has many other features worth noting including analytics and reporting, tools to increase accuracy, and a services team that works with companies interested in custom solutions.

When I talk to companies about using mechanical turks, many are still unaware^†† of what they even are, and most don't quite know how to use them. In our work, we routinely use turks to build machine-learning training sets, and for tasks that require the levels of accuracy that algorithms are unable to deliver. Thanks to companies like CrowdFlower, it's now really easy for companies to dip their toes, and experiment with integrating mechanical turks. And with the launch of their Give Work iPhone app, companies can simultaneously opt to provide income to workers in developing countries.

(†) We are users of CrowdFlower's mechanical turk platform.
(††) Actually nervous laughter is a common response!

As a follow-up to my recent post on the Top Grossing Apps list on iTunes, I examined three lists highlighted in the app store: the Top Paid, Top Free, and Top Grossing Apps. Believing that many users scan these lists, developers covet a spot on any of these Top 100 charts.

In my previous posts, I've highlighted that Games is the largest^† category, accounting for about 20% of unique apps. The graphs below show that the gaming category has a much larger share^†† in each of the three Top 100 lists:

68% of the Top Paid, 67% of the Top Free, and 50% of the Top Grossing apps were Games. Other categories that had disproportionate share of apps in the Top 100 rankings include Social Networking, Photography, (and to a lesser extent) Sports, and Utilities.

In contrast, three of the five largest categories (Books, Travel, Education) were severely underrepresented in each of the U.S. iTunes Top 100 Charts.

(†) Size of a category is measured in terms of unique apps.
(††) Data for this post was for the two weeks ending 10/4/2009. I consider an app as being in the Top 100, if it was listed among the most popular (free, paid or grossing) apps, sometime during those two weeks.

For each decile, I calculated the MEAN price of the Top 100 Apps over the 2 most recent weeks. Notice that for the most recent week, the MEAN price for each decile^† of the Top 100 Grossing apps is more than $5. In contrast, none of the deciles for the Top 100 Paid apps had a mean of $4 or more. There isn't much of a relationship between rank and price although there was a slight downward trend in the price of the Top Grossing apps over the most recent week: except for the blip in the 5th decile of apps ranked 41-50, the top deciles tended to have higher MEAN prices.

The same situation holds when one looks at MEDIAN price during the most recent week: each decile of the Top Grossing apps had a MEDIAN price of $3, while no decile in the Top 100 Paid apps had a MEDIAN price of $2.

Unique Apps by Category: About two weeks ago, the U.S. iTunes store crossed 90,000 apps^††. Last week, the Travel and Education categories displaced Utilities, to claim spots in the Top 4 largest categories:

(†) I refer to an app as being in the Top N, if it was listed among the N most popular (paid or grossing) apps, sometime during the given week.
(††) Since inception, 90K different apps have appeared at some point in time. Over the most recent week, more than 85,000 apps appeared in the U.S. iTunes store.

A little over a week ago Facebook reached a major milestone: 300 million active users. The fastest-growth region continues to be Asia, but growth in other overseas regions such as the Americas and Africa have also been strong. Currently reaching only 1% of potential users in Asia and Africa, Facebook has barely scratched the surface in both regions:

Growth in the U.S. remains fastest among those age 45 and older, and the share of those users is higher in the U.S. than overseas. In other regions recent growth tended to be more evenly divided among age groups. One notable exception has been the teen group in Asia, which grew over 80% in the last 12 weeks.

Of the 300 million users, how many are actively using Facebook right now? (For the rest of this post active means not just logged in, but actually engaged.) By treating the previous question as a Fermi problem^†, I can probably derive a decent estimate. First, I assume that the average fraction of people actively using Facebook at any moment, equals the fraction of time an average Facebook user is active on the site^††. Without access to any usage stats, I'll throw out the following guesstimate: a typical Facebook user spends 4 hours per month (or 48 per year) actively using the site.

Depending on how accurate you want to be, there are 1.6 to 6 million people actively using Facebook right now. If the average Facebook user spends considerably more than 4 hours per month (actively) using the site, the estimate would be much higher than a 1.6 million. I do have an escape clause: in classic Fermi problems, being within a factor of 10 is considered acceptable.

(†) Increasingly popular in the business world, Fermi problems have long been staples in Physics (and Math) departments.
(††) In other words, if the average Facebook user spends 1% of her time actively using the site, on average 1% of all Facebook users are actively using the site at any given moment.

Since we began tracking Facebook demographics in late May, weekly growth has held steady, usually in the low single-digits on a percentage basis. More importantly, it's fair to say that the company has successfully expanded overseas. With close to 128M users, the share of U.S. users is down to around 30% from 35% in late May:

Over the last three months, Facebook has added members across all regions, with the strongest growth coming from Europe, South America, and the Middle East/North Africa:

In Europe, growth has been especially impressive in Italy and Spain. I'm not sure when the Italian translation of Facebook launched, but soon after, Italians started signing up in droves. The (crowdsourced) Spanish translation was completed within a month and launched in early 2008. I've read reports that users in Spain have used the site to connect with long lost relatives in Latin America. Venezuela, Argentina, and Uruguay were Facebook's fastest-growth countries in South America. In late May, some Radar readers were highlighting Facebook's growing popularity in Venezuela, Argentina, and Chile.

I don't have any particular insight into how Facebook is growing in the Middle East and North Africa, but the company has added lots of users in Tunisia, Morocco, and Turkey. (I encourage Radar readers from the region to share their thoughts in the comments.)

Having grown up in Southeast Asia, I've been detecting more interest in Facebook among friends in the region. But for now Facebook still lags Friendster and Multiply. In fact Facebook has far less users in all of Asia than users from Canada! Similarly, the U.K. has more than twice the number of Facebook users than all of Asia. Facebook has to contend with homegrown social networks and slightly different online habits: Asian internet users spend more time on gaming and instant messaging. But even with their relatively small user base and amidst a competitive environment, Facebook is growing in Asia (they added 1.5M users from the region in the last 12 weeks).

Another interesting tidbit about Facebook's recent growth, is that the fast-growing regions discussed above are adding teens (13-17) and college-age (18-25) users at a faster rate than North America.

With a commanding share of college-age users in its home country, U.S. growth has been strongest among working age users (26-59). I was expecting stronger growth in the teen market (13-17), but teens remain the slowest growing group in the U.S.

The Gender split has persisted: Females now outnumber Males, 51% to 44%. In late May the Female to Male split was 41% to 34%. The share of users who decline to state their gender dropped from 24% in late May to 5% in early December.

That Females so outnumber Males may surprise people. While the Female/Male distribution has persisted over time, there is quite a bit of variation across regions. The Middle East/North Africa is the only region where Male Facebook users outnumber Females.

Taking a cue from Raven's recent post announcing the 10,000 iPhone app milestone, I decided to update some charts from earlier posts on the U.S. iTunes app store. First, the weekly growth in the number of apps was slower in November: the number of apps grew less than 10% on a weekly basis for all of November. During the last week of November, there were close to 9,800 unique apps, 22% of which were free.

The average price of a Top 100 paid app continued to decline, falling to a little over $2.60 in the last week of November:

Since high-priced top-sellers actually inflate the MEAN price, I created an alternate chart using the median (the decline in the MEDIAN price is even sharper). The corresponding price distribution continued its downward shift as it becomes harder for high-priced apps to crack the Top 100 paid apps list.

Having more than doubled over the last two months, Gaming remains the largest category accounting for a quarter of all apps. The fastest growing categories were Education and Lifestyle. Medical is the newest app category and as of the end of November there were over 80 medical apps, the 10 most popular of which were free. Among Game apps, Racing, Music, and Sports were the fastest growing Game sub categories.

There has been a slight increase in the proportion of Games priced at 99 cents or lower:

Finally, I computed the share of free apps by category and found that some of the smaller categories have a higher share of free apps. Social Networking apps tend to be apps designed to help users access social web sites from their iPhone, while News apps do the same for news/media sites. In both the Social Networking and News categories, the Free outnumber the Paid apps.

I didn't revisit my previous analysis of the top-selling apps, but I suspect not much has changed since my post at the beginning of November. Among the items I do hope to cover in the future is an analysis of app publishers.

One of my favorite sessions at the recent Web 2.0 summit was Tim's half-hour conversation with Shai Agassi, the CEO of Better Place. Better Place aims to make electric cars widespread ("the electric car as the de facto standard") by addressing major issues that have held back electric vehicles: affordability and convenience.

In a relaxed conversation with Tim, Shai described an electric car industry that resembles the mobile phone business. Just as telecom companies sell mobile handsets at a discount if one is willing to commit to a contract, their subscription-based model will allow consumers to purchase an electric car at the fraction of the normal price. Car owners will pay additional fees based on the amount of miles they drive and the type of car they choose to own. To support their subscribers, Better Place will also build extensive networks of charging spots and battery exchange stations. They will build the first "Electric Recharge Grids" in Israel and Denmark.

Prior to starting Better Place, Shai was a president at software vendor SAP. The interview briefly touches on IT and enterprise computing.

[NOTE: Web 2.0 summit videos are available on YouTube.]

At the Web 2.0 summitt, John Doerr mentioned the high number of games available in the iTunes App store, and wondered whether the iPhone's potential as a gaming platform is being underestimated by Sony and Nintendo. His interest stems from KP having funded a company that develops free and paid games available through the iTunes store.

I decided to pull together a few charts that give an overview of gaming in the iTunes store. The share of game apps has grown steadily and hovers around 25% of (the now more than 6,000) apps available:

As the largest category, Games has several sub categories. The number of apps available in all but three of the gaming sub categories doubled over the last 8 weeks:

Finally, Free games only accounted for 19% of all games in the most recent week and close to three-quarters cost $1.99 or less:

I previously looked into the Top 100 Paid apps (henceforth known as the top-sellers) and found that their average price has been declining. In this post, I'll examine which iTunes categories^† are producing the most top-sellers. In terms of number of unique apps, all the categories have grown rapidly over the last two months, with Sports, Education, and Entertainment posting the highest growth rate.

In the absence of download information, the popularity rankings are the best available proxy for "sales": frequent appearances on the Top 100 Paid Apps list is a good indicator that an app is selling well. Having access to data from around the launch date of the iTunes store, I was able to focus on apps with longer track records (at least 60 days old). The number of days an app spends on the Top 100 list varies across categories:

Of the categories that produce top-sellers, Music was particularly impressive. On average, the 15 Music top-sellers analyzed were on the Top 100 list on 30 different days. (The one Weather app appeared in the Top 100 on 58 different days.) I used duration to see which categories had more proven top-sellers. Solid top-sellers have appeared on the Top 100 Paid apps list on at least 20 different days, while Marginal top-sellers have appeared in less than 10. The Marginal top-sellers have a short-term burst in sales, and are hits for only a few days.

Some categories tend to produce Solid top-sellers (e.g. Music), others produce mostly Marginal ones (e.g. Productivity).

Besides overall sales, Solid and Marginal top-sellers differ in terms of sales velocity (number of days it takes to land on the Top 100 Paid Apps list). On average, a Solid top-seller appears on the Top 100 list the day it launches. It takes a typical Marginal top-seller 5 days to achieve the same milestone. So if an app lands on the Top 100 list quickly, it's also more likely to appear on the list frequently.

With an estimated 7.5 downloaded apps per device, iPhone users are happily experimenting and searching for useful software. While the Top 100 list is a convenient shortcut, savvier users in search of apps do so within categories. Someone in need of a mortgage app will need to navigate within the Finance category. As this short list of top-sellers illustrates, the Top 100 list is a bit like the Billboard Hot 100: it's heavy on the pop side.

(†) While apps can be listed in more than one category, the vast majority (≈ 97%) are listed in only one.

With the recent launch of the Android app market, I thought it would be a good time to give a quick update on the iTunes app store. For now, there are only a handful of apps in the Android market. Developers will be able to start uploading their apps this coming Monday. In the meantime, the rival iTunes app store keeps growing: there are now well over 5,000 apps.

To put the growth in number of apps in context, I decided to look at our data on the Facebook and Myspace platforms. Unfortunately, we got off to a later start in tracking Facebook and Myspace so the graph below corresponds to growth well after their launch dates. While not an exact comparison (data for the iTunes App store is from launch), the resulting graph is still impressive:

I had an earlier post where I talked about the various categories of apps: nothing major has changed on that front. I will do an updated post on categories and other app store topics, but for now I'll focus on the Top 100 paid apps. Since iTunes highlights the Top 100 Paid apps, making it onto that list translates to free exposure.

The Top 100 rankings change constantly throughout the day. They are based on "popularity", but I haven't been able find details of how the rankings are computed (and how often). In any given week applications cycle in-and-out of the Top 100. On average (over the last four weeks) about 136 different apps spent time in the Top 100. But in recent weeks there has been a small decline in the number of apps that made the Top 100:

The average price of a Top 100 paid app has been declining steadily. In mid-August, the average was around $4. Recently, the average price of a Top 100 app was down to about $2.80.

The decline in average price is not just due to expensive/outlier apps dropping out of the Top 100, the corresponding price distribution has been shifting downward slowly over time. It has become harder for high-priced apps to crack the list. Other notable characteristics of Top 100 paid apps:

• On average it takes around two weeks after launch before an app cracks the Top 100 list
• An app can be in-and-out of the Top 100 list over several weeks. On average, an app is on the list on 4 separate calendar (Monday to Sunday) weeks.

Revenue is harder to estimate without number of downloads/installs. If one had download data on enough apps, then a statistical model could lead to estimates for the rest.

We will continue to track the iTunes app store and post interesting trends here on Radar. The app store has clearly intensified interest in the iPhone. Apple sold over 6.8M iPhones last quarter, and Q4 revenues for the iPhone were up 583% compared to the same period last year.

With the U.S. now accounting for only about a third of all Facebook users, we are starting to see a gradual shift away from its original demographic of college-age users (18-25): 46% of all users are 18-25 years old, down from 51% in late May. The number of users in the 18-25 segment is growing, but at a slower pace than the other age groups. Among the major Facebook age segments, the fastest growing are teens (13-17) and young (26-34) to middle-age (35-44) professionals, with the growth in teens driven by non-U.S. markets. Also note the strong growth in the much smaller 45-54 and 55-59 age groups:

In the U.S., 51% of Facebook users are 18-25 years old, down from 59% in late May. But when one looks at other large and/or fast-growing Facebook markets, the share of the 18-25 age group is less than 50% in most of them:

With the iTunes App store now over a month old, I decided to look closely at data from the U.S. store over the last three weeks. While sales numbers are not publicly available, Apple publishes overall as well as category-level rankings. There are currently just over 1,800 (paid and free) applications in the App store, double what it was three weeks ago. Games is the largest category with about 500 applications (roughly 27% of all apps), up 87% from three weeks ago. Puzzles, Arcade, and Board games are the three largest Gaming subcategories:

The fastest-growing category, Education, more than tripled over the last three weeks.

The average price per paid app is around $5.50, with 94% of apps priced at $10 or less. Prices vary considerably by category with expensive apps skewing the average price in a category: a single application priced at $449 drove up the average price of Finance apps to more than $22. Excluding the top and bottom 1% priced apps, the average price of an iPhone application is about $5.20. Similarly, by removing the top and bottom priced app in each category, we get a more reasonable estimate of the average price per app within a category (click here for details).

The Book category is comprised mostly of ebooks and while there are over 150 such "apps", it was the only category not represented in the Top 100 rankings:

In contrast, more than 1 in 10 of all Music apps were among the Top 100 Paid Apps:

Looking beyond the Top 100 paid apps to all paid iPhone applications, the best-performing categories (in terms of popularity) are Music, Weather, Navigation, Lifestyle, and Entertainment (click here for details).

On average, app providers have slightly over one app each, with 25 (out of the close to 1,100) providers accounting for about 21% of all paid apps:

Most providers had 0 or 1 app listed in the Top 100 Paid Applications, with the following exceptions: Hottrix, Pangea Software, Inc., Phase2 Media, telience.com, and Electronic Arts all had 2 apps in the Top 100 list. For now, the cohort of web developers who dominate the Facebook application platform have been unable to make similar inroads in the iPhone platform. Perhaps it's time to brush up on Cocoa and Objective-C?

With the Facebook Developers conference slated for later this week, I thought it would be a good time to give a brief update of a previous post on Facebook demographics. What follows are recently published number of users by country and region, along with growth rates for select regions and countries. Over the last four weeks, the fastest growing regions were South America, Central America and the Carribean:

While Facebook grew double-digits in Asia it did so from a relatively small base (approx. 3.7 million users), in a region with hundreds of millions of potential users. Of the countries in South and Central America, Chile is worth highlighting (up 67.5% from four weeks ago). As several Radar readers predicted, Facebook has grown steadily in Chile where it now has over 2.2 million users (around 14% of the population). In other parts of the Americas, Hi5 and Orkut remain the largest social networks:

Looking closely at the top 30 countries, a few European countries have grown more than ten percent over the last four weeks (France, Spain, Germany, Italy), with France having the most number of users (approx. 2.5 million). Skyrock remains the largest social network in France. Norway saw a decline but is still home to more than a million Facebook users. We will continue to track how Facebook is doing vis-à-vis other leading regional social web sites and whether their disputes with other companies affect their growth rates.

As far as recent trends in the Facebook app platform (the subject of this week's f8 conference), we have detailed reports (here and here) on the subject. At the last Graphing Social Patterns conference, Roger Magoulas provided highlights of our most recent findings. The number of published apps continues to grow steadily (to over 32K) but total usage remains flat. Besides the fact that the top 10% of apps account for 98% of total usage, aspiring Facebook app developers should know that only about 6% of apps average at least 500 active users per day:

(For specific tips on how to launch and build successful Facebook apps, consult this O'Reilly Radar Report.) Finally, as I noted in a previous post, the most popular applications on the Myspace platform continue to account for slightly less users than their Facebook counterparts.

With several hundred applications now available in the iTunes App store, I decided to consider alternate ways of gauging interest in the platform. Using MarkMail, one can quickly scan thousands of mailing lists and restrict the results to those related to software development. Based on the number of posts to (MarkMail) mailing lists, Linux-based alternatives generate considerably more email chatter than the iPhone:

Staying with the previous metric (posts to mailing lists), there does seem to be growing interest in the iPhone among developers. Since the launch of Android (November 2007), the number of iPhone related messages has grown at a faster rate than those for its competitors:

Other online tools suggest growth in the number of job postings that mention the iPhone. But while a majority of the most recent iPhone related job postings were posted by Apple (making the recent growth in job postings less impressive), Android jobs postings came mostly from outside Google.

For now the launch of the iPhone puts the spotlight on Apple's App store and platform. The reality is that the mobile landscape is evolving rapidly and with Android yet to launch, the previous numbers will change dramatically over the next months. We will continue to monitor developer interest in the different mobile platforms using a variety of indicators.

Yet another option lurks, one already familiar to web developers and users. At last weekend's Foo camp, I attended a session on the mobile web and left convinced that with access to the right hooks into mobile devices, web developers can deliver equally cool apps through mobile browsers. Which mobile platform are you most excited about?