Lies, damned lies, and further analysis

A week ago I signed up to StatCounter in order to get some statistics about my site’s visitors. The results are interesting, especially the mobile browsers count.

I signed up for a free account in order to test their mobile browser detect, but quickly extended that to a paid subscription for my entire site.

Believe it or not, but I’ve mostly done without statistics in the past ten years, except for a short period in 2008 or so when I used a bad tool that even managed to mess up its own table-based layout. Until last week I had no clue how many people visited my site and with which browsers, but now I’ve rediscovered it’s fun to know.

The best

In the past few months I found myself returning to StatCounter time and again because it has the best (or rather, the least bad) mobile browser detect out there. So why not use it for my own site in addition to a general market overview?

At this point the following conversation invariably ensues:

Someone
I should warn you that such stats are usually unreliable. Take great care when using them.
Me
I use what I can get. Any data is better than no data.
Someone
At least compare it to other sources.
Me
Such as?
Someone
[... crickets ...]

Ad networks don’t count. They’re skewed towards the platforms they sell most ads on. (Besides, AdMob was the only one, and it has stopped publishing stats. That comes with being taken over by Google.)

Please avoid repeating this conversation in the comments. There’s just no source comparable to StatCounter.

Visitors

I created separate stats for QuirksMode and QuirksBlog. I’m very glad I did, or I wouldn’t have noticed the mobile patterns.

QuirksMode.org, excluding this blog, draws about 22,000 unique visitors per day on a weekday; about 9,500 per day in the weekend. The vast majority of them, about 70%, open just one page that (apparently) contains the technical information they need, and then surf on.

On QuirksBlog that figure is 75%. The total visitor count is uncertain; I have to take my publication schedule into account here. On the whole, I’d say my ramblings and musings draw about 2,000 visitors per day on weekdays I don’t publish; and maybe about half that number in the weekends. Slightly less than 10% of the QuirksMode count, in other words.

However, my organising article proved rather popular, so I got exactly 4,200 unique visitors on Monday, when I published it.

The Sunday stats are skewed because Smashing Magazine tweeted my State of mobile web development, part 3, and it turns out such a mention on the worst day of the week is good for roughly 2,000 hits.

I’ll revisit my QuirksBlog stats in the future, when I have a better idea of the averages.

Browsers

Which browsers do these visitors use? It will surprise nobody that IE use is well below average here, and use of other browsers above average. QuirksMode first:

  1. Firefox: 50%; overwhelmingly 3.6
  2. Chrome: 23%; overwhelmingly 6
  3. IE: 17%; of those 65% used IE8, 22% IE7, 9% IE6, and 3% IE9
  4. Safari: 6%; overwhelmingly 5
  5. Opera: 3%; overwhelmingly 10.6
  6. All others, including mobile: 1%

Then QuirksBlog; and note the subtle differences in the desktop browsers and the large mobile difference:

  1. Firefox: 43%
  2. Chrome: 24%; more Chrome 7 than on main site
  3. IE: 14%; more IE9 than on main site
  4. Safari: 8%
  5. Mobile: 5%
  6. Opera: 3%
  7. iPad: 2%
  8. Others: 1%

Wow. Why these large differences?

Obviously, the lower IE share and higher Chrome and Safari share can be attributed to the fact that the average QuirksBlog visitor is more deeply steeped in the web than the average QuirksMode visitor.

In that light the lower Firefox share is also interesting. For a while now I’ve felt that Firefox has passed its peak and is going into decline (I suppose I’ll have to get back to this in a future article). Do these stats prove my visitors agree with me?

Time will tell; for the moment the sample is too small. But this is one more thing to monitor.

The mobile side of things

Let’s drill a bit deeper. I followed the mobile stats intensely from the very first hit, and I noticed something interesting. The following numbers exclude the iPad.

Here is QuirksMode:

  1. Safari: 44%
  2. Opera Mini and Mobile: 24%
  3. Android: 19%
  4. Nokia: 3%
  5. BlackBerry: 3%
  6. Bolt: 2% (Bolt is an independent WebKit-based Opera Mini-like browser)
  7. Other: 6%

Now QuirksBlog; in absolute numbers about the same amount of visits, so the relative number is about ten times as large as on QuirksMode:

  1. Safari: 64%
  2. Android: 25%
  3. Opera Mini and Mobile: 7%
  4. BlackBerry: 2%
  5. Other: 2%

Wow, what a difference. What happened here?

Smashing Magazine did.

As I said Smashing Magazine mentioned one of my articles on Twitter. From that moment on my mobile stats changed radically. First of all mobile as a whole grew tenfold from about 0.5% to 5%.

Secondly, where before the tweet Safari was only narrowly ahead of Opera and Android, after the tweet its traffic share grew to the current numbers. Android and BlackBerry kept up, while Opera and Nokia dwindled to nothing.

Apparently, this Twitter link triggered a disproportionate amount of mobile traffic, consisting especially of Safari and to a lesser extent of Android. The other mobile browsers didn’t play a role any more. In addition there was disproportionate iPad traffic.

I’m deliberately not going to draw more conclusions because the dataset is so tiny. All in all we’re talking about only 900 or so mobile hits on QuirksBlog, and 1,000 on QuirksMode.

Still, better analysis of mobile stats in relation to mentions on Twitter (or, I presume, other social media sites) could find some interesting differences in traffic market share.

My theory is that Safari and Android usage, as well as mobile usage as a whole, and possibly tablet usage, is significantly higher on sites that get lots of traffic from social media.

Let’s see if future analysis bears it out.

And let’s see if this pattern, if there is any, continues. After all, the masses of Nokias and BlackBerries will hit the web at one point.

StatCounter problems

In order to use StatCounter as a tool for such research, some things should change. I’d welcome some kind of open API or even a simple form where I can paste a browser string and retrieve StatCounter’s identification. Update: this already exists.

In addition, the mobile browser detect should be changed.

StatCounter made some weird decisions when it comes to iOS devices. It splits them up in iPhone, iPod Touch, and iPad, and counts the first two as mobile devices, while the iPad is added to MacOS Safari.

iPhone and iPod Touch show up separately in their mobile stats, and that makes absolutely no sense since from a web development point of view they’re the same device. Worse, it confuses people (hell, it confuses me) because the mobile browser overview seems to show Opera well ahead of iPhone, while it’s only ahead by the narrowest of margins once we include iPod Touch. Similarly, BlackBerry appears to be the second-most-popular mobile browser, while it is in fact third.

As to the iPad, is it a mobile device or is it a ... well ... non-mobile device? My current answer is Yes.

From a UI perspective it’s hard to tell whether the iPad belongs on the mobile side or not. Both positions are defensible. The large display definitely argues for non-mobile, but the fact that it’s much easier to pull from your pocket and use than a laptop argues for mobile.

Technically there’s no doubt: the iPad runs Safari for iOS, and not Safari for MacOS. I want to know the total traffic share of Safari for iOS, so from a development point of view the iPad should be put in the “mobile devices“ category. Which only shows that that category may not make as much sense as people think.

And what about Opera? Annoyingly, StatCounter does not split up Opera traffic into Mini and Mobile. I think that Opera Mobile is about 5 to 10% of the Opera total and Mini the rest, but that’s really only guesswork.

StatCounter, please make the following changes:

  1. Lump together iPhone and iPod Touch.
  2. Split up Opera Mini and Opera Mobile.
  3. Split up Nokia into its component browsers and OSs (this is tricky; I’m willing to help)
  4. Add version numbers. This is especially important on BlackBerry, where OS6 comes with a WebKit browser and the lower OSs don’t. Web developers need to know.
  5. Go through the list of currently-minor browsers (such as Dolfin for bada) and make sure they’re detected properly. At least one of them will become important to detect in the future.
  6. Allow two-dimensional browser/OS reports: I’d like to know how many Firefox users are on Mac, and how many Mac users browse with Firefox.

If you do that you’ll become a truly valuable mobile browser analysis tool and I’ll happily pimp you.

This is the blog of Peter-Paul Koch, mobile platform strategist, consultant, and trainer. You can also follow him on Twitter.
Atom RSS

I’m speaking at the following conferences:

(Data from Lanyrd)

Categories:

Monthlies:

Comments

Comments are closed.

1 Posted by kl on 14 October 2010 | Permalink

You can have 2-dimensional stats in Google Analytics, and with a bit of hacking with GA filters and Custom Segments, you could filter out noise from their browser detection.

GA recently published PHP script for JS-less detection too.

2 Posted by Glenn Glerum on 14 October 2010 | Permalink

I am curious about the future blackberry stats, how will they change when the tablet comes out?

What defines a mobile device? The definition for mobile is "moving or capable of moving readily". As you stated the iPhone is easier to pull from your pocket. But how big is the smallest laptop? 8inch? does that count as mobile as well?

3 Posted by David Storey on 14 October 2010 | Permalink

You can detect mobile devices and report them to StatCounter by going to http://gs.statcounter.com/detect/

It's hard to say whether iPad should be counted as mobile. It certainly isn't a mobile phone (but neither is iPod, so you could argue that should be in a different category too). Its usage pattern is certainly different. There isn't the always on aspect (like iPod) as it is a WiFI device (unless you have the 3G model) and is something you are probably more likely to use in the same context as a laptop (in the coffee shop, sofa, when sitting down) rather than a phone were you use fully on the move, such as when standing on the train or whatever. Its usage pattern is probably somewhere in the middle of a laptop and a mobile.

4 Posted by AlessandroS on 14 October 2010 | Permalink

If you needed good stats... Why don't you just analyze you server logs? It's the most accurate one, as it is not influenced by whether JS is on or off, etc etc...
There are daemons that do that automatically on the server...

5 Posted by Dave Hulbert on 14 October 2010 | Permalink

Have you given http://piwik.org/ a go? It's self-hosted and open source, so you could add any browser UAs you like. The ones it currently gets are listed here: http://dev.piwik.org/trac/browser/trunk/libs/UserAgentParser/UserAgentParser.php

6 Posted by AlastairC on 14 October 2010 | Permalink

Google Analytics recently added a 'mobile' section under visitors.

You can cut it by OS (although confusingly it separates iPhone/iPad/iPod), but perhaps more useful would be the cut by browser.

I get results like:
Safari, "Mozilla Compatible Agent", BlackBerry9700 (and other variants), NetFront, LG, HTC_HD2_T8585 Mozilla, SAMSUNG-GT-S5620 etc.

You can display both to cross-reference them. (The Mozilla agent seems to be from iPhone and SymbianOS, so may be Opera mini.)

Worth a look perhaps?

7 Posted by ppk on 14 October 2010 | Permalink

Where are Google Analytics' public stats?

Exactly.

That's why I won't use it. Besides, I once tried to find the actual code I need to insert into my pages to see if it is hackable, but I couldn't find the information at all. Too confusing.

8 Posted by Aeron on 14 October 2010 | Permalink

I would like iPod and iPhone to be separate. Remember that you can also track stats from apps with embedded webviews. In that scenario I would like to know which of my users can take advantage of SMS, GPS and other telefony goodness...

9 Posted by Steve on 14 October 2010 | Permalink

AlessandroS

Re server logs being best.

What about pages that are viewed from caches elsewhere than your server?

10 Posted by pakoiy on 14 October 2010 | Permalink

"Lump together iPhone and iPod Touch.
Split up Opera Mini and Opera Mobile."

So basically, make the iBrowser look better, and Opera worse? :)

11 Posted by gossi on 14 October 2010 | Permalink

I guess, putting the iPad into mobile is wrong. The iPad opened up a new category, the tablet. A bigger mobile device and we are going to see more tablets in the wild over the next years. So, it doesn't belong to desktop nor to mobile.
That's what I want to see as a developer. iOS Safari, for browser market share either on mobile or tablet and the same for screen resolution and design objectives.

12 Posted by Rob on 15 October 2010 | Permalink

@kl do you have a link for that PHP script?

13 Posted by R. Bemrose on 15 October 2010 | Permalink

"Someone
[... crickets ...]"

That crickets part should be "parse your web server access logs using a web analytics program."

Now, the problem here is that you need to find a stats processor that actually separates browsers by mobile and desktop types in addition to the browser distinction types. Last I checked, packages like Awfful (a fork of Webalizer) didn't yet do this. Having said that, Awfful has user-defined browser groups.

Google Analytics is powered by Urchin, a proprietary web analytics program whose creators were bought up by Google in 2005. Unlike Google Analytics, Urchin can scan your logfiles directly.

Urchin is also available as a standalone program, although it is unclear at this time where you can get a license; Google claims that you can get an Urchin license from a third party, but their website doesn't specify which third party.

14 Posted by Nicolas Chevallier on 17 October 2010 | Permalink

I think mobile stats will be better in the next years. iPad has been released just a few months ago so it's normal that there's some question about that new revolutionary device. Perhaps it's not a mobile nor a not mobile device? a new category seems to be a good alternative.

15 Posted by Steven Tomlinson on 30 October 2010 | Permalink

iPad's are mobile.

You aren't tethered to a power supply, a desk or a location when using them. Though most importantly, they are designed to be used when changing location. They just aren't as portable as iPhones.

Giving an iPad a different category leads into all semantic problems. Where does a laptop fit in these categories? What category does a game console go in?