Searching for Microsoft market share will yield a variety of statistics, opinion, and debate. Common sense tells us that specific figures for user base size are accurate, and are therefore more informative. Below the surface of numbers and marketing is a more complex story, however.
Software sales by Microsoft count three sources: boxed software on merchant shelves and online stores (CompUSA, Best Buy, etc.), installs on computers before sale (referred to as OEM - Original Equipment Manufacturer), and volume sales through vendors. The first two are very inaccurate in determining the number of Microsoft software users. Counting software sold through stores includes boxes sitting on merchant shelves and storage rooms, i.e. sold by Microsoft but unused. Store purchases also don't account for the software a customer is already using. The new software can overwrite the old software or the old software may still be put to use. Sometimes store purchases are made because OEM disks are lost, causing a user to be counted twice. The majority of computer vendors who sell their hardware with software installed primarily use Microsoft. It's been determined in federal court that Microsoft has used its desktop monopoly to arm-twist vendors into signing contracts requiring their sale of Microsoft sofware with their hardware. Using hardware vendor sales to count software users proves inaccurate from discounting those who replace the Windows operating system with an alternative such as Linux® or FreeBSD after purchase. It also miscounts many corporations which purchase new computers and put older versions of Windows on them.
Statistics also discount those who purchased Microsoft® Windows or Microsoft Office™ but later decided to uninstall it. None of these non-users are subtracted from the software sold by Microsoft when determining user count. Even attempting to account for these sales which don't apply to the size of the user base will prove inaccurate. A very broad survey of home and corporate users would have to be taken. A valid survey would become useless very soon after it's taken because the share of users is likely to be changing rather quickly.
All of this tends to show Microsoft's user base is smaller than most believe. In addition to the true number of Microsoft users being unknown, market share statistics are even more complicated due to the nature of [open source] software such as Linux. The nature of open source licenses promotes free sharing, which only surveys can count. Download statistics are often used to assist in determining open source market share, but barely tell more than the basic popularity of software packages.
John Lettice, writing for The Register, explains how 2002 license shipment statistics can only be partly used to analyze the server market. Research firms such as IDC routinely ignore anything beyond sales figures. It's simply impossible to account for many scenerios, such as free downloaded software being used in place of purchased licenses, or purchasing a Windows XP license but installing Windows 2000.
A thorough explination of the myths of software market share data can be found in a LinuxWorld.com article: "There are dozens of reasons why people have underestimated how quickly Linux has been grabbing Windows' market share. Windows starts out with a false boost and maintains its illusory market share even as it gets replaced by Linux."
For corporations making the [software development] platform choice, size of the user base helps determine if something is generally found useful and if enough knowledgeable developers can be found. It's nearly impossible, however, to get accurate statistics. Few of the platforms are actually for sale, so market share can generally be ignored. The most useful information comes from surveys.