Fun with Excel

It seems that Pogson has recently discovered the wonders of spreadsheets.

“One picture says 1K words but some provisos need to be added. The absolute numbers here mean nothing except share on the one site. The changes in the numbers are what is important here.”

No joke, Poggy.

Let me give you a lesson here on data analysis, Pog. (Don’t worry – it’s free of charge.) First of all, you don’t just assume correlations between two dimensions without proper justifications. That means, in your case, you don’t just put up a logarithmic graphic without explaining as to why the correlation is indeed exponential. Of course, you can argue that the growth in human population or even the growth in domestic PC ownership in many countries (!) has shown to be exponential, but that doesn’t mean the “attach rate” of any OS – or, in this case, Linux – does indeed follow such a trend. Secondly, r-square is your friend. No, it’s not bulletproof, but it’s nonetheless better than the sweet nothing you show on the graph. Thirdly, drop OO Calc. Seriously, kill it with fire, already.

I am all for good examples, and this is why I’ll take the opportunity here to show you how statistical analysis like this should be done. Hell, let’s go a bit provocative and measure each OS’s attachment directly against that of Windows!

Here’s what I did using the exact same data Pog had for his graph:

• For the sake of the argument, each datum was assumed to be the exact equivalent of the percentage market share of the corresponding OS at the corresponding point in time.

• For each sampled month, all recorded instances of “Windows” are combined to approximate the actual total market share, in percentage, of the Windows series of operating systems.

• The market share of operating systems unaccounted for, in percentage, was approximated for each sampled point in time using the difference between 1 and the sum of percentage market shares of Windows, Mac and Linux. In other words, these “Others” operating systems were assumed to be non-Windows, non-Mac and non-Linux (e.g. BSD).

• For each non-Windows OS, a fraction of 1 is taken by dividing its percentage market share by that of Windows.

Table 1: Monthly counts of OS attachment in fraction of Windows attachment, ([OS % market share / Windows % market share] vs. time; click to see image in original size)

Notice, in Table 1, the most-likely exponential trend in Mac’s growth against Windows. Notice also how I have plotted the trendline linear for Linux – don’t worry, unlike Mac, there is strictly nothing interesting in the r-square value between a linear plot and an exponential one before the third (and last) significant figure, and even given that the amount of difference is pretty negligible (~0.01). Hell, even the category “Others” shows a promising sign of exponential growth after its sharp decline in 2008. In other words, if there was anything to blame for Windows’ negative attachment rate, Mac would be a far, far more likely candidate than “GNU/Linux” and everything you had suggested. Seriously, Pog, what game are you trying to play here?

——-
“Then there was the man who drowned crossing a stream with an average depth of six inches.”
— W. I. E. Gates