????By Philip Elmer-DeWitt

????The so-called market share reports issued every month by Net Applications have long been controversial — mostly because they didn’t actually measure market share (which business people typically express as the number of widgets they sell in a given period divided by the total number of widgets sold).

????What Net Applications did instead was sample data from browsers visiting their clients’ websites and report what percentage came from machines running Windows, Mac, Linux, etc.

????But despite their flaws, we tracked the net metrics firm’s reports because their sample size was relatively large — some 160 million visits per month — and because they offered regular snapshots of broad market trends. They revealed, for example, the rise of Firefox, the decline of Internet Explorer, the failure of Windows Vista to catch on. Their reports were consistent, dependable, and free.

????Until now.

????Net Applications’ reports are still free, at least some of them (mobile market share data now costs an arm and a leg). But their dependability — and perhaps their credibility — just took a huge hit.

????Starting in June the company changed the way it weights its data, giving more weight to page views from countries with large Internet populations that aren’t well represented by their clients (such as China) and less weight to hits from countries like the U.S. that are over-represented in their data.

????Wrenching changes

????The effect was to cause wrenching changes in the results — so wrenching that Net Applications skipped its June report entirely. And on Saturday, when it finally issued its July report, the new country-by-country fudge factors were applied retroactively to all past reports.

????The new market shares for the major operating systems are represented in the pie chart above. To see how different they are from the old, we have to go back to May, the last month for which we have comparable data. The results are summarized in the table below. The major effect of the new weighting system:

????? Microsoft (MSFT) Windows’ share grew more than 6%

????? Apple (AAPL) Mac OS X fell more than 51%

????? The iPhone OS lost nearly 60%

????? The iPod touch — whose rapid growth was the subject of a Net Applications featured report — fell off the chart

????? Java ME — Sun Microsystem’s (JAVA) plaform for mobile devices, barely a blip in previous reports — grew 212%

????Net Applications has retroactively revised previously published data before (see here), but never to this extent.

????In defending the changes, the company argues that they provide a much more accurate view of worldwide usage share statistics. But it also acknowledges that the new method produces “some significant changes in usage share” — usage shares it published with great confidence only a few months ago. The changes it highlights:

????? Baidu – Baidu goes to 9% of global search engine usage. Baidu is on a major growth curve, which is affecting the relative share of all other search engines.

????? Google – Because of Baidu’s growth, Google’s global share is actually going down. This is almost completely due to Baidu and does not reflect the rest of the world.

????? Apple – Since Mac share in the U.S. in significantly higher than the rest of the world, Mac and Safari share drop in the global reports.

????? Opera – Opera goes up to 2% in global reports. This reflects the significant share they have in Eastern Europe and Asia.

????[What the company doesn't mention -- but several readers have pointed out -- is that their operating system numbers are now much more in line with StatCounter's global stats.]