Showing posts with label Statistics are weird. Show all posts
Showing posts with label Statistics are weird. Show all posts

February 01, 2019

Statistics are weird

After losing market share for six straight months, Windows managed a very slight uptick last month, gaining 0.03% to 86.23%. This very slight gain appears to have come at the expense of MacOS and Linux, both of which ticked down slightly, by 0.06% and 0.33%, respectively; ChromeOS gained, though, climbing 0.05% to sit at 0.37% overall, its highest point in the last month.

Windows' slight uptick didn't help its 6 month average decline, which actually worsened as last July's 0.55% increase finally moved out of frame:


This month's 0.03% increase is unlikely to have the same salutary effect in Windows' 6mo average, though -- it's just not large enough to have much of an impact, falling below the ±0.05% threshold for actual significance. ChromeOS and MacOS changes this month were right on the significance bubble, too; only Linux' 0.33% drop was large enough to have a long-term depressing effect on its 6mo average tracking.

The picture gets even muddier when you dig into the different OS versions. Sure, Windows 10 gained last month, from 39.22% to 40.90%, but Windows 7 also gained, from 36.90% to 37.19%; these gains were offset by losses by Windows 8 (-0.06% to 0.82%), 8.1 (-0.11% to 4.34%), and XP (-1.78% to 2.76%). On the Linux side, the generic "Linux" category declined 0.28% to 1.72%, but Ubuntu only declined by 0.03% to 0.68%, while the "Unknown" category gained 0.30% to reach a whopping 0.35%, which sure looks like NetMarketShare had trouble detecting peoples' OS versions this time around.

So, with statistical anomalies all over the place this month, I'm going out on a limb to call this a case of sub-optimal data, and putting a great big asterisk next to the whole mess until next month. If this month's numbers truly do represent some sort of inflection point, then we'll see similar changes in coming months, too; if these are just July-like blips in the data, then we should return to the trend lines of the past 6 months.

Either way, I want more data before I even try to guess at what's happening here.