According to NetApplications, GNU/Linux now accounts for 1.02% of computer desktops. Depending on your outlook, this figure is cause for cautious celebration, or equally cautious dismissal of the operating system's prospects.
Alternatively, you might prefer -- as I do -- to question the statistic's accuracy, and look at other ways to estimate GNU/Linux's presence.
The problem, of course, can never be settled with any certainty. GNU/Linux comes from dozens of vendors and community projects, rather than a single corporation, and, since it requires neither registration nor activation, a single DVD could be the source of half a dozen installations. Under these circumstances, any estimates quickly trail off into speculation whose credibility depends on your own biases on the subject.
Still, if other estimates may be too high, we have good reasons for thinking that NetApplications figure to be inaccurate, just by looking at the statistics provided. To start with, by separating out figures for Android, you could say that NetApplications has actually under-estimated the size of GNU/Linux's market share by .07% -- a small amount, to be sure, but a reminder of how easily such figures can change with your assumptions.
Notice, too, that the figure of 9.73% for the Mac is almost twice the one usually bandied about for OS X. That difference is enough to suggest that NetApplications' figures are skewed in some way. At the very least, the margin of error in extrapolating from a small sampling to worldwide usage must be extremely high (something, to be fair that NetApplications itself doesn't do, just those extrapolating from the data it releases).
Without NetApplications's raw data, its sources of bias are impossible to discover. However, I would guess that GNU/Linux users would be more likely to choose free software website applications than NetApplications's, or perhaps a free online service such as Google Analytics.
In addition, NetApplications is an American company, and its website lists no overseas office. For these reasons, I would assume that its figures are drawn largely from the United States and perhaps Canada, and less so from the rest of the world.
Given that the United States was only ranked ninth in open source activity in the Open Source Index recently published by Red Hat and Georgia Tech (and Canada twenty-eighth), such a bias would seriously question the general applicability of the NetApplications figures.
Given Microsoft's power and influence, few would be surprised that its products are used more heavily in its home country than anywhere else. Yet, despite the North American technical community's frequent delusions, it is not the entire world, nor, these days, even necessarily the most computerized or fastest growing part of it.
The Approach Through Firefox
So how can we find alternative, preferably more accurate figures? Mozilla Corporation employee Asa Dotzler offered two lines of reasoning a few months ago in his blog.
In one, Dotzler accepted that 1% of computer users were on GNU/Linux. In the second, he based his figures on Firefox usage. Keeping his figures conservative, Dotzler assumes that at least 90% of all computers are connected to the Internet. He also assumes that Firefox is the default browser for 75-90% of GNU/Linux users.
Based on direct checks for software updates on Mozilla's pages (a starting point that omits updates through each distribution), he estimates that GNU/Linux users represent 2-5% of Firefox users. Given 85 million tracked users, Mozilla estimates its install base at about 250 million, a number he says that Mozilla has found generally accurate over the years.
From these assumptions and figures, Dotzler concludes that GNU/Linux users number somewhere between 5.5 and 16 million. If you accept Internet World Stats figure of just under 1.6 billion computers, then Dotzler's answer is between .5-1.5%, a figure close enough to NetApplications not to seem too controversial.
However, a number of those responding to his blog post suggest that his estimate is on the low side. The suggestions are based largely on instinct, but they are right: unless I missed something, Dotzler apparently forgets as he calculates that not all GNU/Users use Firefox. If you accept that 90% do, then the real figures are 6-17.6 million. If you prefer the lower estimate of 25%, then the figures become 7-21 million, or as high as 1.5%.
And what about those who only install Firefox through their distribution's package management system? I can think of no way to estimate their numbers. However, given the widespread reluctance I have encountered from desktop users to install compressed tar-files of the type that Firefox offers for download on its site, I would be surprised if they were not the majority, or even two-thirds.
So, depending on the figures you choose, the GNU/Linux install base on the desktop could easily be 2-5% by this line of reasoning.
The Indirect Approach
A few weeks ago at COSSFest in Calgary, I heard a talk on this subject by KDE's Aaron Seigo. Here, I recreate his logic from memory and imaginative extrapolation, so my apologies if I get any parts of it wrong. The point is the general argument, anyway.
Seigo began his estimate by a story about how a scientist in the 1940s supposedly estimated the force of an atomic bomb by measuring how far it displaced scattered scraps of paper. He tells this story to emphasize that he derives his estimate from indirect indications, rather than hard figures or direct observation.
Observing that North American trends are not always in keeping with those of the rest of the world, Seigo points out the countless indications of growing GNU/Linux adoption: data points like the growing number of computer manufacturers offering pre-loads, and the operating system's success on netbooks.
He talks about cities and government departments in Europe adopting GNU/Linux, of 53,000 labs in Brazil deploying the operating system to 52 million students, of the dominance of Red Flag Linux in China, and similar success stories throughout India and South America, pointing out the rapid growth among populations equal in size to North America and often larger.
Look outside North America, he says, and the generally accepted figures are almost all too low, distorted by the Windows-dominated perspective of the United States and Canada.
At the end of his presentation, Seigo was reluctant to give a percentage, but, when I pressed him, he suggested that 8% was probably a bottom figure, and 10-12% the likeliest. Some of the audience seemed politely skeptical, but, even if you half his figures, his examples go a long way towards suggesting that North Americans consistently under-estimate GNU/Linux use because we live in a Windows-distorted environment.
Some of his scenario seemed too rosy. For example, he did not discuss how, so far as the Chinese government is concerned, one use of Red Hat Linux seems to be to force Microsoft to lower its license fees. Nor did Seigo address claims that many GNU/Linux boxes distributed in Brazil soon have copies of Windows installed on them.
Still, even if you doubt Seigo's exact figures, his presentation did suggest that, whatever the user statistics are now, they're about to rise sharply.
Choose your logic, choose your percentage
The wide variation in the figures that I've mentioned only emphasize the difficulty of determining user numbers. Apply a different set of assumptions, be stricter or laxer in your estimates, and the numbers can vary wildly.
All the same, exploring the possibilities is still worthwhile, if not for the original purpose of establishing hard figures. Instead of giving an exact number, running through the logic, the assumption and the biases behind the numbers is a useful reminder that the objectivity of numbers is often just a myth.
Are GNU/Linux desktops as rare as NetApplications suggests? As widely deployed as Seigo would have them? We have no way of knowing which is closest to the truth.
But at least Dotzler and Seigo, free software advocates that they are, explain their reasoning, instead of just offering a number with the illusion of objectivity that media services and columnists will pick up without asking questions.