All this shows me....
...is you can't rely on these stats.
It's the first Monday of the month, so time for us to have a look at desktop operating system market share again! Keen Reg readers may recall that last month it looked like peak Windows 7 was behind us. This month's data supports that hypothesis: Statcounter has Windows 7 dipping a little to 54.13 per cent and Netmarketshare …
"Care to explain why?"
I can't speak for the OP, but from the article: "Windows 8.1 is up from 10.04 per cent in January to 10.49 per cent in February on Netmarketshare's numbers, and from 14.27 per cent to 14.78 per cent according to Statcounter."
So two groups try to measure the same thing and they come up with figures that differ by about a third. Back when I was in science classes, if I had two measurements that differed by that much then I'd expect to have my knuckles rapped if I quoted either to 2 decimal places. The correct way to interpret these figure is to call them "experimental proof that at least one of them hasn't got the precision they are quoting".
I do trust these stats to be accurate records of two different outfits' atempts to measure OS use and have no difficulty with the number of decimal places quoted.
Think about the problem, they cannot take a sort of photo-finish snapshot of the whole internet at 10:00 hours GMT; I imagine it's not technically possible and in any case it would skew the results towardsr European users. So they have some measuring procedures that don't overly favour certain times and places and basically add up very large numbers of detected OSs.
Nothing wrong with that and the results are intriguing.
@Uffish - The issue isn't whether they can accurately count the number of hits their servers get. It's whether that count is an accurate representation of the entire population of the Internet. They're extrapolating their own limited sample sets of data to the Internet at large. The fact that the numbers from these two companies differ so much from each other shows at least one, and probably both, sets of samples are in fact not representative.
One major problem will be internationalization. How many Chinese, Russian, Indian, Indonesian, etc., web sites versus American and UK web sites are sampled by either of these companies? The proportion of Windows XP in use is believed to be much higher in the first set of countries than in the latter ones. How many users on business PCs using Windows XP are visiting shopping sites versus the proportion of Windows 8 users on home PCs? That will skew the numbers as well.
At best these numbers will show general trends. Reading meaning into hundreths of a percent though is both wrong and pointless. All we should be looking at is the general trend line.
ok, once more. Try to get your head round the idea of "an accurate representation of the entire population of the Internet". Suppose it were possible to get an accurate snapshot, or 3D animated GIF, of such a thing - what use would it be?
The answer - no use at all. No-one has a use for such a global figure. It is simply an amusing concept for a coffee break.
Hence my untroubled acceptance of the results kindly reported in el Reg.
@Uffish - "Suppose it were possible to get an accurate snapshot, or 3D animated GIF, of such a thing - what use would it be?"
Well, companies who are in the business of selling hardware, software, and services associated with Windows would like to have accurate market share numbers so that they can compare those with their own sales figures to see how they are doing.
Is XP use declining while Windows 8 increasing? If so, then if the sales figures for the Windows 8 version of your product are flat, then your own market share figures have a problem. If sales are going the other way, then you can tell if your sales force are doing a good job, or if they're simply benefiting from increased sales of Windows 8.
For people working in the technology business in general, they like to have a good idea where the market in general is going, so they can focus their personal energies on things that have a future.
For web browser usage figures (another thing people look at), it tells you if you should be taking a closer look at how well your web site works with particular browsers. You don't want to use just your server stats, since you can't tell if certain browsers aren't showing up as much because people are no longer using them, or if it's because visitors got fed up with bugs in your web site that your Mac using iPhone toting web design consultants are oblivious to.
On the other hand, extending the decimal places too far to the right in the numbers will make you see trends that aren't there. Was a bump in Windows XP in the December figures due to people using old PCs at home over the Christmas holidays, or was it just random noise? Look at all the news stories which try to "explain" each twitch of the figures based on speculation that these numbers are accurate. These are the data points upon which a lot of speculative extrapolation is based, and we need to be skeptical about them because the data to support it just isn't there.
Whether or not 0.07% or 7% is significant is a difficult question that depends on the amount of data and how accurately it is measured. Your suspicions about the "margin of error" are therefore baseless. They might be reporting on samples in the millions where tiny fluctuations can be measured.
Having said that, it is the Register, and I'm certainly not going to invest or gamble any money based on their reporting. To me, the real focus of the Windows XP topic should be that Microsoft's threats and blackmail aren't working perfectly, and as too often happens, the Register is off chasing some codswallop.
And they can't stick it in some kind of VM while they work out what to do with it?
These stats are measuring where XP is running. It's not going to make any difference whether it's on physical or virtual hardware. Of course if you have to run XP then running inside a VM is probably a good idea but still not a panacea. You might not be able to isolate the applications from the outside world and even if you can it's still possible to suffer a breach if the damage occurs within the VM itself.
Walk through most airports en-route from check-in to the plane and see how many PCs you pass which are running XP - I did it yesterday and my count was 4 (at least 2 of which were sat beside dot-matrix printers, which probably also says something about "if it ain't broke don't fix it").
I travel most weeks, and the count at most airports is usually about the same, which when you consider how many gates most airports have these days quickly tots up to a lot of XP installs...
Not just you. I'm wondering how far people are rolling back Windows to get away from 8. Past Vista apparently.
On a serious note, I know of 3 people (home users not business) who have killed Win 8 and installed Win XP for compatibility issues and/or comfort zone issues. So it does make me wonder.....
could also be people who have moved from XP to a new Windows, and have now moved off Windows entirely to another OS
Not quite, certainly some have gone that way, but it's curious that the upward trend in Windows XP is almost matched by the downward trend in the others.
If it were users going from Windows 8.x to non-Windows, I'd expect the Windows XP trend to remain flat and the bump to show up in either the MacOS X or Linux trend lines. (Of course, there's *BSD too, but they don't show a trend line for that family.)
Not too difficult:
Replace old (XP) PC because it's really not getting the job done anymore.
New PC/Laptop comes with Win8 flavour.
Cursing and frustration ensues.
Look into the comments about Win8, and the Promised Land of Win10 around the corner.
Get Handy Cousin to install XP until Win10 arrives **
You'd expect Win7, but XP simply has more "This one will work" codes in the wild.
(** or swap in old harddisk, or any variety)
Still in the shrink wrap - I have a whole spreadsheet with 200+ never used OEM HP and Lenovo codes for XP from machines which were never booted into Windows (or had an enterprise license with a different key installed at first boot). So I can always install one if I ever need a windows system (least likely as with the exemption of an occasional work laptop my house has been Microsoft OS free since 1997)
Until last week it had been possible simply to download Win 7 ISOs from Digital River. These would install and authenticate normally using an OEM key; well, certainly this worked with HP machines and licences. Now the Digital River downloads have been stopped and only those who have a key from a full retail copy are able download the ISO, after first registering their key with Microsoft. Those who bought a machine with Windows 7 already installed are instructed to contact the manufacturer if they wish to obtain an ISO, even though their key is valid.
Far from loosening its grip, Microsoft has made it impossible for the majority of users to download a Windows 7 ISO.
Shame the handy cousin didn't just install Classic Start Menu...
Last I checked, Classic Start Menu was merely cosmetic, it didn't magically make your Windows 8 installation magically sprout drivers or run older software.
We have to use Windows XP at work for some things: namely I have to because of some Windows CE 5 devices we support that require ActiveSync 4.5, and some of my colleagues do because the Rockwell Automation software requires it for running PLC simulators.
Everywhere else it's either Linux, Windows 7 or Windows Server 2008r2, the former two being the majority.
Oh dear. So a company publish results that are to do with the IT world in general. El Reg publishes what the company said, and also states quite clearly:
"even when one takes into account the less-than-stunningly-accurate method both market watchers use of observing traffic as it hits web servers."
Plus, no spelling errors that I noticed, so a solid win I thought.
That is a good point. What we can postulate is that the readership of El Reg is more technically orientated unless there are more masochists or lost carrion bird fanciers than we could assume would visit. This means that we expect a skew compared to the internet as a whole.
This means we have a smaller pool which targets specialists. So a reasonable assumption is that a smaller pool for one website will not be reflective of the netmarketshare pool, unless they are only operable on The Register.
Now, it would be possible to check the benchline of OS breakdown for the Reg and it normally wouldn't be unusual to see if there is a corrosponding drop in stats for 8.1 and rise for XP, but at the same time, XP is more likely *not* to be used amongst IT professionals apart from those locations where they are forced to in an enterprise situation and theregister isn't blocked.
So we have three assumptions here:
1. Netmarketshare is more generalised than The Register visitors.
2. The Register visitors are more likely to be in IT service businesses/departments
3. Netmarketshare doesn't just show results from the The Register.
Reasonable taken separately, but combined that then starts to increase liklihood of error.
Not remarkable a conclusion, maybe, though I thank you for making such a claim on my behalf - I don't often get such praise. But it certainly is a reasonable set of assumptions to make that any corroboration The Register visitors make is only ever going to be meaningless considering the size of the dip/rise in question, it would just be too small a comparison even if there were a similar spike/dip.
Then there is the possibility of a DOS attack from a botnet of compromised XP pcs on one particular site - that could produce an alteration in stats of that magnitude, theoretically.
Welcome to statistics.
Collect another badge of fail. Your assumptions 1 and 3 are the same and number 2 is very similar to them both (if A is more generalised than B then B must be more specialised than A. You only qualify this by indicating where the specialisation may be). It doesn't really matter because the assumptions are only relevant in a hypothesis that you are going to test empirically. Which you can't because El Reg doesn't provide the numbers. This is logic, we haven't even got to the statistics.
Furthermore, you might want to get out your dictionary and look up corroboration. El Reg's numbers don't have to be the same as either of these two services to provide corroboration. However, in the past the numbers have *not* been corroborated by data collected by Akamai to which I have repeatedly referred.
"Collect another badge of fail. Your assumptions 1 and 3 are the same and number 2 is very similar to them both (if A is more generalised than B then B must be more specialised than A. You only qualify this by indicating where the specialisation may be). "
Assumptions used in a conjecture do not necessarily have to be a reasonings without overlap - one and three are restating the same thing but the language used shows the approach to each one. It may be said that by stating the assumption from two viewpoints clarifys what is being said.
But it *was* sloppily stated. Can't downvote myself but I will upvote you on that.
"t doesn't really matter because the assumptions are only relevant in a hypothesis that you are going to test empirically. Which you can't because El Reg doesn't provide the numbers. This is logic, we haven't even got to the statistics."
Very true. The fail badge is worn for this alone.
"Furthermore, you might want to get out your dictionary and look up corroboration. El Reg's numbers don't have to be the same as either of these two services to provide corroboration."
I know they don't have to be the same and that wasn't what I wanted to convey. What I should have said is the numbers will be considerably smaller overall so any corroboration indicated could be considered an anomaly - you would really need another data collecion over a larger pool of websites - like you say, Akamai's own findings.
I am not defending the result of netmarketshare, I just don't think that El Reg's stats would be meaningful in determining whether or not XP increased market share. But, as you say, without evidence of the figures then this conversation is arguably pointless, although I welcome your well reasoned arguments. It could be that El Reg's figures show an *large* upsurge in XP takeup which would certainly shoot my expectations down.
I am not defending the result of netmarketshare, I just don't think that El Reg's stats would be meaningful in determining whether or not XP increased market share. But, as you say, without evidence of the figures then this conversation is arguably pointless, although I welcome your well reasoned arguments. It could be that El Reg's figures show an *large* upsurge in XP takeup which would certainly shoot my expectations down.
You might be surprised at what the numbers reveal if only we could see them. Unfortunately, we'd then have to do some real statistics to point out another glaring problem with the numbers that El Reg reports: desktop is steadily losing marketshare to mobile devices. This skews desktop stats to older machines and also makes them more volatile as the sample size decreases.
I've had a look at Windows 10 using the Technical Preview release ISO which is free if you 'sign up' with an email address. It looks like a magazine for teenagers and has some aspects that are confusing for me regarding how you arrange the desktop and get shortcuts to work. I'll admit that I tried to make it look like XP so I'm sure I'll be happy to continue using Linux MINT (Mate) which is very easy to make look like XP at the desktop GUI level.