For the second month in a row, NetMarketShare are reporting that Windows 7 is gaining market-share faster than Windows 10. It's difficult to know exactly what this means, and for who, but one possible explanation is that users upgraded to Windows 10 and rolled back to 7 in significant amounts. It will be interesting to monitor the next couple of months, now that Windows 7 is no longer available at retail, to see how its market-share shifts. Then, a few months after that, we'll need to see how Zen and Kaby Lake, which are not supported by Windows 7 and Windows 8.x, changes that further.
I'll now spend the rest of the post discussing statistics… because I can visualize the comments.
NetMarketShare records browser identification strings from partnered websites. As you would expect, there's a bit of controversy regarding how accurate their numbers are. Some of this criticism is simply wrong, usually misunderstanding how small a truly random sample needs to be to converge to the same ratios you will see in a large sample. Just a thousand truly random samples can get you within a few percent of hundreds of millions of people. Studies like this, if they are truly random, have plenty enough data to get a very precise ratio.
A valid concern, however, is whether their pool of websites under- or over-represent certain groups, especially when you attempt to make comparisons on the order of a hundredth of a percent. NetMarketShare claims that they try to get a global representation, including government websites, and they correct their traffic based on the CIA's per-country statistics. Still, it's good to question whether the group of people you are trying to investigate are represented by NetMarketShare's traffic, and how their limitations lower your effective precision.