Developer Analytics reports are based on a unique comprehensive media measurement system combining surveys, and third-party panel based datasets with our internal statistical algorithms.
Despite our prolific stage presence directly on social media properties, there is always some information that slips by. In fact, you might be surprised at just how much activity exists "off the radar." As of 2008, there are ~27,000 applications in the Facebook product directory, but there are over 200,000 applications off the Facebook product directory. Although we don't include the majority of these applications (many are inactive, incomplete, or simply blank pages) we still keep an eye on non-trivial amounts of traffic driven to lurking social media properties.
Panel data acquired from third party partners provides usage patterns of hundreds of thousands of randomly selected users. These data points provide the information we need to extrapolate highly accurate page view counts even for social media properties that don't include our direct measurement code.
What's the simplest way to find out what apps people are using, what music they like, which movies they watch, and what books they read? You ask. You ask often, you ask randomly, and you ask a lot of people. Survey data acquired from users interacting with CPA networks, partner applications, and our own distribution network comprises the high level information we provide on audience profiles. Not only that, but trends within demographic subsections of social media properties can be used to further hone page impressions extrapolated from panel data by weighting user behavior by demographics.
We regularly poll developers and analyze revenue trends within our own applications and those of our partners to create accurate prediction algorithms for CPM, CPA, and virtual goods monetization potential. Our equations take into account return users, churn rate, daily uniques, application age, and page multiplier to determine what applications currently are, or could be making with each monetization technique. Our trend equations are derived from hundreds of REAL data points, and produce realistic approximations of monetization potential. Larger applications have the capacity for larger branding deals, but there are insufficient data points in the industry so far to provide predicted data for branding opportunities.
We're fairly confident with many of the numbers we report on our site. Why are we confident? Manual data integrity checks that we use to refine and harden our extrapolation equations. We compare our predicted usage data with real data provided to us by other developers and our own applications to ensure the validity and accuracy of our reports, and investigate any major discrepencies. We enjoy understanding every peak and valley in our charts, and checking our data manually as often as we can helps us guarantee a high level of quality for our premium subscribers.