iOS Ebb and Flow

I’d like to fill in the picture I began in my recent post on Android fragmentation by examining the changes in version distribution of the other major mobile operating system: iOS.

Unfortunately, this was no quite as easy as it was for Android. Unlike Google, Apple does not publish the version distribution of its user base. What we do have, however, is a number of developers who have published the version distribution within their own apps, and if we can collect a large enough sample it may be feasible to use these in lieu of direct vendor-supplied data.

Using 50 data points from different developers, we can indeed build an image of version distribution over time for iOS just as we did for Android. Note that these are bundled into major releases. Grouping these into the major releases reduced noise and also matched the groupings more closely to those I used for the Android post — while it may not be a direct Apples-to-Apples comparison, it is likely the best we can do.

Versions by Date

The data is obviously imperfect as it comes from multiple sources, but the trend lines have R-squared values of between 0.8 and 0.97 (R-squared of 1 indicates that the line perfectly fits the data), so we can feel confident in their use. It’s interesting to see the symmetry as one iOS version rises to power and the previous version falls, a pattern that was not strong in Android as multiple version coexisted simultaneously.

As we did with Android, let’s index these versions to the same starting point so that we can see how quickly versions gained and lost share relative to the number of weeks after their launch (same colors as the previous chart):

Versions by Weeks After Launch

If we include the Android versions (again indexed to the same starting point), we get an extremely telling image of the differences between the two platforms:

Version by Weeks After Launch with Android

iOS 5 captured approximately 75% of all iOS users in the same amount of time it took Gingerbread to get 4% of all Android users. Even more astounding is that 15 weeks after launch iOS 4 was at 70% and iOS 5 was at 60% while Ice Cream Sandwich got to just 1% share at the same age. If there were any question as to whether iOS had a less fragmented ecosystem than Android, the past two charts provide a fairly definitive answer.

Some folks have told me that it is unfair to compare iOS and Android on this metric because iOS is effectively just three devices (iPod Touch, iPad, iPhone), whereas Android is a multi-manufacturer ecosystem with dozens of devices. This line of thinking is extremely frustrating to me. Developers and users don’t care that the two platforms aren’t the same. Users want the most recent features and security updates, and will demand them either directly (by complaining) or indirectly (by making a different purchasing decision), and developers want a unified base to minimize testing. Android apologists can list off the differences between the two all day long but it doesn’t change the fact that more versions with smaller share is worse for, at the very least, developers and users.

A Model for iOS Version Uptake

I would like to try and provide some value to (iOS) developers in the form of a model to estimate the proportion of the iOS user base on a given version after its launch.

To do so, I took every data point available that showed the percentage of all iOS users on the most recent version at the time the data was generated (so, this data will be an average of iOS 3.X, 4.X and 5.X). These were plotted against the number of weeks that had passed since the version was introduced.

Percent by Weeks After Launch

The data isn’t a perfect fit (R-squared value of about 0.7), but there is a fairly clear upward trend with gradual levelling out. Based on the average uptake pattern, it takes less than a week for iOS versions to reach 25% share of all devices, and a little less than six weeks to reach 50%. On the other side of the coin, older versions tend to lose an average of 3% of their installed base per week. Though I have been performing my analysis using major version groupings, this pattern seems to apply equally well to the point releases as it does to the major releases, and we can assume that it will only accelerate now that Apple has introduced over-the-air updates for iOS.

Let’s again take a look at how this compares to Android versions:

Time to Percent Milestones

iOS devices have, on average, reached 10% version share 300 times faster than Android versions, 30% share 19 times faster, and 50% share 7 times faster.

In a way, I think that iOS buyers are paying to be on the cutting edge of software. Android OEMs have been one-upping each other on the hardware front (the Android spec race has reached almost ridiculous proportions), but this is a shallow, easily-duplicated strategy. An ecosystem that has been developed instead with a software focus affords many advantages that are not easily mimicked: ease of development, users being able to learn about apps and the OS from friends without the frustration of fragmented device capabilities, and more.

How Many iOS 5 Devices Are There?

As a brief aside: we can use the numbers above to come up with an approximate number for developers looking to target only iOS 5 users as their addressable market. If we assume 100% year-over-year growth for both iPhone and iPad sales, we can estimate that about 170 million devices are now on iOS 5. Comparing this to the largest single installed base on Android, Gingerbread, which runs on about 112 million devices, shows us that iOS makes up for a lower absolute number of devices by presenting an extremely large and unified base for developers to market their apps.

Thanks to the following individuals/ companies for providing iOS version share data in a publicly accessible manner: Marco Arment of Instapaper, David Lieb of Bump, David Smith of Cross Forward Consulting, game4mob, Chitika Insights, Flurry Analytics, and Apprupt.

You can follow me on Twitter as @pxldots.

All of the data used for this post can be freely downloaded:

  • Excel file containing all of the collected data and calculations is here.
  • Numbers file containing most of the charts is here.
  • PDF containing the images of all charts used is here.