Slashdot Mirror
StatCounter Blasts Microsoft's Claim About IE Still Being the Number 1 Browser
An anonymous reader writes "Do you remember when Microsoft tried to claim that Internet Explorer was still the most-used browser by accusing StatCounter of using a flawed methodology? Well, StatCounter has just posted a response that walks through a number of errors and omissions in Microsoft's reasoning. They (rather politely) explain the importance of sample size, discuss the value of page view counts versus unique visitor counts, and explain the difference between their methodology and that of Net Applications."
It matters to the people who like to use these stats when they show negative things about Microsoft. But when the same source publishes stats that show Linux has less than 2% marketshare they decry the source as being untrustworthy.
237 out of 237 Microsoft employees recommend Internet Explorer
Join the Slashcott! Feb 10 thru Feb 17!
That seems about right for linux desktop market share. Who is claiming it is not?
Now claiming that includes all the servers out there can't be right.
I was, at least, interested in the bit about why they use Page Views instead of Unique Visitors. My initial reaction would have been to side with Microsoft with the Unique Visitors metric, but StatCounter makes a great case...
- Person opens IE on a machine (for whatever reason) and uses a site that's part of their network. Let's call it five pageviews.
- Then they close IE and use Chrome or Firefox for 500 more pageviews, to every other site, for the rest of the day.
Now using Uniques, you'd show that person as an IE user. Or at maybe you'd 50/50 it. Both methods poorly represent that person's browser usage than the total pageviews by browser. It's not perfect, but it does make sense.
Their youtube video makes it quite clear, and it's good that they did this.
What's the actual truth?
You see, SC comes up with a moderately intelligent article that does seem, in the face of it, to address the points Microsoft addresses.
And yet, virtually anyone who administers a public website can tell you that SC's original figures are complete crap. IE most certainly is the most popular browser right now. And Chome is third place. Not second. Definitely not first.
SC can continue to push this ludicrous crap if they want. But their figures are laughable, and they'd be better off figuring why than writing snippy retorts to anyone who points it out.
You are not alone. This is not normal. None of this is normal.
If you do any web development, these browser stats control how much testing and effort you have to spend supporting various browsers. IE unfortunately is difficult to work with as the "standards" they choose to implement often work differently than the standards that Firefox, WebKit and others use.
For every Intel server sold (irrespective of operating system) there are simply thousands of PCs sold. It further doesn't help that the vast majority of servers are generally not used to surf the web, while most PCs are... which will further deflate the number or Linux servers counted by services such as StatCounter.
So, from a unit count perspective, I hardly doubt that including the number of servers running Linux in the count is going to make a noticeable dent in the Linux market share statistic.
Server's have a larger market share of value spent (and to a lessor degree of CPU cores sold), than the units sold would suggest.
Stat Counter probably counts all devices and there is a ton of these things called Android that uses Chrome.
Yes, but this is a relatively unusual workflow. Even as a power user, the only time I ever launch a different browser is when I'm testing a website to make sure it works in another browser. Unless you use some particular site that works correctly only in a particular browser, most people simply do not use multiple browsers on a regular basis, and sites that work correctly in only a single browser should be excluded from this sort of statistic anyway. So basically, the benefit they're claiming is better precision for the 0.5% of people who have intranet sites that are IE6-only and then forget to switch browsers when visiting a web page initially. It's lost in the noise. More to the point, because those people forgot to switch browsers, they don't really care which browser they use, so as far as web developers go (deciding how important it is to support a given browser), they really are 50-50 because your site could support one or the other, and those users wouldn't really care which.
By contrast, most people use only a single browser. Thus, for the 99% case, if there are differences between the page count stat and the per-user stat, this tells you that people who use certain browsers tend to look at fewer pages. This may be an indication that the browser sucks, but it also may be an indication that your site does not support that browser well enough, or it may be an indication that the sorts of users who use that browser are simply too busy to spend time browsing a lot of sites. Thus, the two numbers provide significantly different information, and the question of which one is more useful is largely dependent on why you are asking the question. If you are trying to find out how many people will hit your site with a given browser, the per-user stat is more useful. If you are trying to figure out which browser is more likely to have people browsing around your site and looking for products, the per-page-hit stat is more useful. Understanding the differences between the two metrics can also help you better tailor your site to the sorts of users that browse it using different varieties of browser.
Check out my sci-fi/humor trilogy at PatriotsBooks.
Based off the link you posted above, from an article submitted three years ago?
People change. People's opinions change, especially with articles like this that illuminate the different methodologies and reasoning. Different people exist on the website.
Not to mention that there could be entirely valid reasons why the StatCounter stats could be entirely correct in this case and still be flawed in the determination of the OS share.
I'm not sure why you're trying to create doubt and controversy here.
I vote based on politicians' actions, unless contrary to my preconceptions. Often wrong, never uncertain. #iamthe99%
I certainly believe statCounter as it it logical and things with NetMarket do not make sense like IE 6 going up 15% usage last January?? The statCounter shows smooth graphs with less variation and I agree that no one outside of slashdot runs linux.
I stopped using Linux in March 2011 after Gnome shell, Unity, and all the new browsers hit 6 week release cycles for security and bug fixes. Linux lost out for me and millions of users.
However, it kicks ass on the server. NetMarketshare does not even show Linux either.
I do not understand the obsession of Linux beating MS. I used to be in that crowd when Windows 98 was so problematic but those days are LONG gone. Linux never does just work if you do updates as the lack of ABIs with drivers and config files changing becomes a nightmare when you have 3k apps installed. I am seriously not trolling but just stating my experience as someone will say it works fine for me etc. Ubuntu is not kind with my laptop and older desktop.
Android and IOS are more consumer oriented and people running Linux servers should not be browsing websites on them.
http://saveie6.com/
Most people I know DO use multiple browsers. Hell, my girlfriend, a college student majoring in food science (not quite a tech field) has Safari, Chrome, and Firefox on her macbook -- and uses all three. Personally I ALWAYS have at least two browsers open -- Chromium and Firefox. Use it to have multiple sessions of the same website open; or because some pages load faster in one browser than another; or because it's easier to remember certain tabs are in Chromium while others are in Firefox, rather than having multiple windows or trying to scroll through dozens of tabs. And not a single one of those wouldn't apply to a casual user -- I'm not talking things like testing or debugging websites...though I commonly switch and add browsers for that too.
I think you may also be underestimating the cases of people switching over for a specific website, and then leaving that browser up for a while. I just graduated from PSU and can tell you that on their course management system (ANGEL -- which is used by several other universities) you can't do certain tasks from Firefox (like sending emails); other tasks won't work on Safari (don't remember specifics, since I don't own a mac.) So any student there who prefers Firefox or Safari will probably end up switching browsers frequently -- I know some people who have to do that multiple times a day. And this is generally from their home computer, so it's likely they'll continue surfing with that browser until they close it. My highschool's website didn't work on certain browsers. My current work webmail and portal system won't work on certain browsers -- and it's a freakin IT company! Point being, there's never been a time in my life when I DIDN'T need to switch browsers on a near daily basis, for reasons that have nothing to do with being a 'power user' or web developer. And I don't know a single college student -- business major, agriculture, engineering, whatever -- who doesn't have a preferred browser.
Of course, all of this does miss the point that they should be able to take that into account in their statistics. If you view a site 5 times in IE and 95 times in Firefox, add .95 to Firefox and .05 to IE. Statcounter has the data to do that...though I'll admit I haven't read this thing fully -- someone please enlighten me if they explain a reason they couldn't do that.
My issue with NetMarketshare is they only look at 40k websites and they have a tiny amount in China. They weigh it to try to makeup but the number in China is so small that variations are included in the data.
For example, Arstechnica said IE 6 usage is up by 15%! Then it mysteriously goes back down by an equal amount. Then goes back up next month by 11%. Most IE 6 users are in China and not in the US (contrary to popular belief on slashdot that corps make it up). So if you have such a tiny sampling size in China who owns most of all IE 6 users you will see strange statistics.
Now if you make an English website that means the statistics are useless as your boss might thing this means American websites and a surge in IE 6 usage means you need to work 70 hour work weeks for the next month downgrading your site to work with such a small marketshare of users.
Statcounter admitted they screwed up and fixed Chrome pre-rendering if you look at their site starting last month it takes this effect into account and Chrome lowered a little bit.
http://saveie6.com/
The reality is that those numbers don't really matter if you already have a website.
You can easily run stats on YOUR OWN WEBSITE and get the browser breakdown that you should be worried about.
For one of my primary sites, all version of IE beat out Firefox or Chrome. When split apart, Firefox and Chrome are 1 and 2, with IE8 coming in third.
And now that I think about it, knowing who is first or second is pretty much irrelevant. What matters is the percentage of users who are still using browser version that suck to support. So really, what I care about is where my IE7 and IE6 usage is, and at what point is it okay for me to walk away from those users.
The world moves for love. It kneels before it in awe.
but then there is also the embedded browser factor - I open my newsreader app and that counts as a IE page view, or I open my OSS dev tool and that counts as a webkit view.
Nowadays its not easy to really get anything other than a broad estimate of browser usage.
Well, if you want to be pedantic, for ever PC sold these days, there are simply thousands of embedded devices sold - a market which is pretty much owned by Linux and *nix variants.
As for how many browse online? Hard to say, though most if not all the Android installs do.
Quo usque tandem abutere, Nimbus, patientia nostra?
"Then what do they have to track you?"
Your unique system+browser configuration?
Slashdot.
"Microsoft Sees Linux As Bigger Competitor Than Apple"
"Microsoft CEO Steve Ballmer showed a slide showing, from Microsoft's internal analysis, that Linux client use is clearly ahead of Apple's."
When the fuck did Slashdot become exactly equivalent to Microsoft? I thought this was still a second-hand tech article aggregator and instant DDoS machine.
Let q be a radix > 1. I am in ur base-q, killing 10 d00ds.
Who is claiming it is not?
Microsoft.
In February 2009, Steve Ballmer of Microsoft presented a slide based on Microsoft's research; while it showed no figures, the pie chart depicted Linux and Apple as each having roughly 5–6% of home and business PCs.
"I've got more toys than Teruhisa Kitahara."
Hard to say how many Windows PCs are being retired every day, but given most corps work on somewhere between a 3-5yr lifecycle, it'd probably be a significant proportion of Microsoft's installed base.
W8 will probably accelerate the downwards slide.
"I've got more toys than Teruhisa Kitahara."
I have to take issue with StatCounter's claim that their data is inherently better because they have 3 million sites in their sample vs 40,000 in Net App's sample. Ask any statistician (I'm not one, but I do fiddle with stats from time to time) and he'll tell you that the only situation in which having more than 40,000 data points (and Net App had 40,000 sites of data points, meaning many millions of page views) can make any difference is if you're trying to tease out extremely subtle differences.
Regardless of the total size of the population you're trying to estimate, you only need a relatively small number of samples to get a given degree of certainty that your hypothesis is not invalidated by your data. This is why you see nationwide polls that only ask 2,000 people out of 300 million Americans -- because the math shows that's all you need to achieve a +-3% margin of error with a 95% confidence interval, and note that you can achieve the same accuracy with the same number if you randomly select 2000 people out of the seven billion on the planet. The margin of error depends on the sample size, not the population size. Once you're up to tens of thousands of samples, the margin of error is miniscule, and upping that to a few million samples doesn't appreciably improve your accuracy.
What does matter, a lot, is that your samples are randomly-selected. And the fact is that neither StatCounter nor Net App have a very good story to tell there. StatCounter's larger sample size may possibly help by getting a slightly larger cross-section of the web, but I doubt it. Both companies measure only a tiny slice of web usage, so complete coverage is a pipe dream, and both have way more than enough data to achieve highly accurate estimates -- if the data is well-sampled, which it isn't. If it were, their estimates would be identical to several decimal places.
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.