Android fragmentation is a topic that swells up now and again, but most discussions lack evidence and are typically laced with vitriol and “fanboy” accusations. I would like to work through a quantitative approach (beyond just version distribution numbers) that can be used as a measuring stick for the severity of fragmentation over time.
Luckily, a fair amount of data on this topic is available and even more can be reasonably estimated. Google has published monthly version distribution figures since the beginning of 2010. They have also made periodic announcements about the total activations to date and the device activation rate, which Horace Dediu of Asymco has compiled here.
Let’s start with the basics: Android version distribution over time.
While it may be interesting to see how versions wax and wane over time, it’s a challenge to pin down exactly how “bad” the fragmentation is at any given time. In order to develop a model that measures this, we have to define what makes a particular version distribution better or worse than another. There’s no right or wrong way to do this, but the model I would propose is based on two tenets: the more handsets on the most recent version, and the less divided the remaining installed base (aside from those on the most recent version), the better. Using these two factors I built a formula that provides us with a value of how “bad” Android fragmentation is; it can theoretically go from 0–12.5, with higher numbers indicating “worse” fragmentation. The results of applying this formula to historical version distribution yields the following chart:
From a historical perspective, Android fragmentation was actually at its lowest level recently (December 1, 2011). Since then we have returned to high fragmentation, but this is typical after a new version is released. We can also begin to see how slowly the fragmented Android ecosystem improves; extremely low early adoption of the new version, in conjunction with multiple older versions having a large share of the non-current-version devices, make for up to six months of highly fragmented user base.
I consider the above analysis fairly one-dimensional and I would be remiss to stop at that. A logical next question should be, “is Android fragmentation improving over time?” I would rework this question to be, “are Android versions rolling out more quickly over time?” One way to answer this question is by examining the percentage of handsets on the most recent version.
Note that each new line segment starts on the first date that market share was registered by Google, while the cutoff for the version “zones” is the date the version was publicly available, so there is some overlap.
It looks as though the newest version is making its way into the market more slowly over time, as represented by the slope of the lines. This effect is even more pronounced when we look at the percent of devices on a version as a function of weeks after launch:
It’s clear that Gingerbread has disseminated into the market much more slowly than either of Froyo or Eclair. In fact, it took Gingerbread about 17 weeks longer to reach a version distribution milestone (10%, 20%, 30%) than its two predecessors. While it is too early to fairly judge ICS’s trajectory, it certainly appears to have started at a slower pace than did Gingerbread (more on that later). A seemingly endless string of devices entering the market with Honeycomb and few older devices being upgraded to ICS makes it unlikely that we will see the Android version distribution improve in the near term.
In fact, I feel it is conservative to state that 2012 will be the year of Gingerbread, at least in terms of version distribution. Gingerbread appears to be on the verge of peaking as a percentage of the total devices in use, but it took Froyo over 6 months after reaching the peak of its relative distribution to be overtaken. As I’ll go into next, Gingerbread is still adding devices 14 times faster than ICS.
Market share certainly isn’t the only factor we need to consider. Importantly, a developer won’t be put off by low market share if the absolute number of devices is significant. I would argue that ICS’s low market share would mean little if the absolute number of ICS devices were large. Here are two illustrations of the total units in use by version, assuming a two year life for devices:
Three versions have significant installed bases at present: Eclair (15 million), Froyo (54 million) and Gingerbread (114 million). As one might expect, Honeycomb has not performed well: only 6.6 million Honeycomb devices estimated in use, which we can compare to the 15.5 million iPads sold just last quarter.
ICS’s slice is almost imperceptibly small, but the Android pie is quite large, so it’s hard to tell if this is a significant user base. One last chart can solidify our thinking on the rate of new Android version adoption and how attractive the newest version may be for developers and users: estimated weekly unit changes by version.
This final chart certainly doesn’t bode well for those who would hope for a more unified Android user base on ICS. Only 200,000 devices are being added per week, approximately the same speed as Froyo and Eclair entered the market. ICS was introduced into a much larger user base than previous versions, so we would expect a more rapid uptake through a combination of higher sales and software upgrades from phones running older versions.
If you want to play around with, rework, examine or critique my data, please feel free. You can use the two files below to do so:
All of the charts are included in a .pdf document here.
The Excel file with all of the data and calculations is here.