Stats Weenie

I can't reveal details of the sources of these numbers - they're sites I work on professionally - but they are both e-commerce sites aimed at the British market, and the figures are based on sessions for the week ending 1st December 2006.

          Site 1    Site 2
IE6        72.1%     74.0%
IE7        14.2      13.6
Fx1.5       5.5       5.3
Fx2.0       2.7       2.6
Safari      2.2       0.9
IE5         0.9       1.1

Traffic levels are such that the figures after the decimal point probably aren't statistically significant enough to be worth the pixels they're rendered with, but I thought I'd bung 'em in anyway.

This reinforces my comments in my last post - IE7 is being adopted very rapidly indeed. Something like 6 weeks after the release of IE7, something like 15-17% of British IE users appear to have upgraded. If you're responsible for sites, and you haven't tested them in IE7 yet, you're missing the boat.

It also hammers home a realisation I came to a couple of months back - for many people (including me) IE5 is no longer significant. It's starting to sink into the background noise.

So, what's been going on since I went silent?

The most obvious news is probably IE7. After a lengthy public beta phase (which itself saw IE7 become visible in stats) IE7 for Windows XP was released to English-speaking users on October 18th. The first batch of other languages followed on November 2nd, with more to follow.

A few weeks after the initial voluntary-download releases, each language will start distribution via Windows' Automatic Update feature. For English, this started on about November 1st.

Mozilla also released Firefox 2 on October 24th. Again, distribution of this release is driven by an automated upgrade tool. [Update: No it's not. Or, not entirely, anyhow. I'm still trying to figure out what the precise situation is]

Therein lies the less obvious and to me, more interesting, piece of news - both of these browsers are now actively pushing major upgrades (as opposed to bug fixes) out to users. The effect is dramatic in these figures from my favourite eastern-european source:

period        Firefox 1.x  Firefox 2.x
17/10-23/10      21.5         1.0
24/10-30/10      20.0         3.1
31/10-06/11      18.4         5.0

Source: Gemius SA, gemiusTraffic, 17/10/2006 - 06/11/2006. http://en.ranking.pl

Allowing for the market share increase over the period (and assuming all the new users went straight to Firefox 2) this means that at least 10% of all Firefox users in that sample upgraded in the space of 2 weeks. I suspect that the week-long sampling period is actually distorting the Firefox 2 figures downwards. Even allowing for some levelling off, that means Firefox 2 will probably overtake Firefox 1 by Christmas.

I don't yet have any similar figures for IE7 that I can publish here (I'll try to get hold of some soon) but I've seen samples from UK-based sites showing IE7 at 4-6% of total users for the first week of November - which is to say, within days of the Automated Update push starting. Before the push started, IE7 was at about 2%. Again, that's a minimum of something like 3-4% of IE users upgraded within a week of release.

When IE6 was released, we had to wait over a year for it to pull level with IE5. I would be surprised if IE7 doesn't overtake IE6 by the end of January - which is to say, within 3 months of release. In the grand scheme of things, that's INCREDIBLY fast. We haven't seen people changing browsers (even if only upgrading to new versions of their existing browser) this quickly since the height of the IE vs NN browser war, which was nearly a decade ago.

After years of a mostly stable browser landscape (and the rise of Mozilla/Firefox was generally pretty slow) things are suddenly moving very quickly indeed. I think the speed of this browser upgrade cycle is going to catch a LOT of people by surprise. I've already seen some major UK high street name sites failing badly in IE7 - I doubt they will be the only ones.

Apologies for the fact that it's gone so quiet around here recently. This is mostly due to real life intruding.

At this point, given that I can't get the content together, it seems more than a little selfish to keep things locked up. So, I've opened posting to everyone, albeit still moderated by yours truly. I'll apply the moderation stick to keep things on-topic and generally civilised, but that's about it. Questions are just as welcome as self-contained essays. I'm not totally familiar with the LJ moderation tools, so apologies in advance if things are a bit bumpy at first.

I also have some real posts of my own bubbling under, which should appear here in due course.

A few weeks back, OneStat.com released another of their irregular breakdowns of global market share. At first glance, the main headline figures contain few surprises: Firefox up, IE down, end of story. However, comparison with their past figures show a few wrinkles:

Browser     10/2005  04/2005  02/2005
IE           85.45    86.63    87.28
Firefox      11.51     8.69     8.45
Safari        1.75     1.26     1.21
Opera         0.77     1.03     1.09
Netscape      0.26     1.08     1.11
Other         0.26     1.31     0.86

The October numbers, on their own merits, seem perfectly reasonable, but the changes compared to the April/February figures are little odd. The fall in IE usage has slowed down somewhat, but the growth in Firefox has increased dramatically. Safari, previously steady, has jumped significantly. Opera has slipped quite a bit, and Netscape has collapsed totally. The "other" figure is also jumping about wildly. For me, these changes are worrying, and I suspect that there's been another change in counting methods.

The growth in Safari usage seems plausible; other sources are showing growth in Safari's share (presumably, as OneStat conclude, Safari is displacing IE on the Mac platform) and the disconnect from the February-April growth could just be statistical noise.

More problematic is sudden jump in Firefox's share and the collapse in Netscape's share. Most sources show Netscape at low levels and declining, but nothing like losing three-quarters of their users in six months. Similarly, most other sources show Opera level or growing, not falling. Finally, the jump in Firefox's share is more substantial than shown by others, though the final figure of 11.5%, while high, is not completely unreasonable.

Down at the bottom of their release, OneStat mention some details of their methodology:

Methodology: A global usage share of xx percent for browser Y means that xx percent of the visitors of Internet users arrived at sites that are using one of OneStat.com's services by using browser Y. All numbers mentioned in the research are averages of [last week of October 2005] and all measurements are normalised to the GMT timezone[?]. Research is based on a sample of 2 million visitors divided into 20,000 visitors of 100 countries each day.

(emphasis added)

There's not enough detail here to be certain, but bits of that sound worrying. Compiling individual statistics for each country is great (I'd love to see some of those numbers, for a start). However, if they then compile the global figures by adding together the individual stats for 100 countries, then they are going to be way off. Put simply, some countries are much bigger than others - just adding together the numbers for 20,000 Brits and 20,000 USians will not give an accurate representation of the total, simply because the UK only has about a quarter of the population of the USA.

One hopes that OneStat have thought of this and compensated accordingly. Unfortunately, simply dividing out by the population of each country will probably not be enough, as things will be even more confused by the number of people in each country who have internet access. The US and UK probably have similar levels of penetration for Internet access now, but this was certainly not the case 5 years ago. Such comparisons for say, the US vs (say) Greece are almost certainly invalid.

Similarly, simply averaging the seven days of the week could cause problems. Internet demographics are radically different at the weekend from the working week.

All that aside, they have provided some of their individual country stats, which are unaffected by the geographical concerns, and make for interesting reading in their own right:

Browser     Global   US   Canada   UK
IE           85.45  80.73  78.52  93.37
Firefox      11.51  14.07  16.98   4.94
Safari        1.75   3.55   2.05   0.99
Opera         0.77   0.77   1.67   0.39
Netscape      0.26   0.76   0.68   0.23
Other         0.26   0.12   0.10   0.08

As I've remarked before, large differences between countries are normal. In this case, the language barrier is less significant, with all three countries being primarily english-speaking (though Canada has a substantial francophone population, of course).

The high usage of Safari reported in the US is expected, as the Mac platform is far more popular in the US than the rest of the world. Some of the other differences are a little more surprising, however.

While the exact figures for UK Firefox usage in the UK vary from source to source, most that I've seen to date (for example, the bbc.co.uk figures reported here recently, and XitiMonitor's various surveys) put it similar levels to the US. So to see such a large discrepancy here (14% vs 5%) is unexpected.

Similarly, the comparatively strong showing for Opera in Canada is a surprise. While there are other markets, particularly in Europe, where Opera routinely shows much higher shares (3-5%) I would have expected Canada to follow the US. Shows how much I know.

(Just to prove that I ain't dead, just busy)

The wires are a-buzz with links to an article published a few days ago which analyses the web browser stats for www.bbc.co.uk (just the homepage, not the whole site). It does make for fairly interesting reading, covering the OS breakdown as well as browsers, but if you're reading this, you're probably only interested in one set of numbers, so here they are. I've number-crunched a bit from the raw figures, hopefully without making any horrible assumptions.

Browser                  Share
Internet Explorer 6      78.3
Firefox                   9.7
Internet Explorer 5.x     3.8
Internet Explorer 5.5     2.9
Safari                    2.6
Internet Explorer 4.0     0.9
"Netscape"                0.7
Mozilla                   0.6
Opera                     0.3

The data set is pretty big (about 32 million requests) and appears to be hit-based. The sample covers "one week in September [2005]" (which week is not specified). Unidentified user-agents are excluded from the total, as are those that accounted for fewer than 50 requests.

It's kinda tricky to make any judgements about geographic distribution of all this, but given that these figures are for auntie beeb's homepage, rather than the news site, I suspect that things are skewed heavily in favour of us Brits. Of course, this is a Scientific Wild-Ass Guess, but it's all I've got right now.

If so, then the approx 10% for Firefox ties in well with recent figures published by French stats outfit XiTiMonitor, particularly after allowing for that source's tendency to report high figures for Firefox.

A few months back, replyhazy asked me to predict what sort of impact the just-announced release of Netscape 8 would make on the browser scene. Five months after the Beta and three months after the actual release, I think it's time to check how close I was:

Given the reception of the last major Netscape release (the world is underwhelmed) [...] I think it's likely to sink without trace.

From what I've seen, Netscape 8 is pretty difficult to identify accurately unless you know what to look for in your logs. Under the hood, it's the bastard split-personality offspring of IE and Mozilla. Most of the time, it uses the mozilla gecko rendering engine, but sometimes it uses IE's Trident instead, and it appears to send a user-agent string based on the rendering engine. A quick skim of EWS's weekly log yields the following examples, showing both flavours in action:

Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1) Netscape/8.0.2
Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1; SV1; .NET CLR 2.0.50215) Netscape/8.0.2
Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.7.5) Gecko/20050519 Netscape/8.0.1
Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.7.5) Gecko/20050720 Netscape/8.0.3.1

Few sources are identifying it at all, but here are a couple that do. These are sources with a fairly good record for detecting minority browsers; bear in mind that they have a somewhat different profile of users than most US-based stats sources. I've included earlier netscape releases and a couple of other minority browsers for comparison.

Browser      ranking.pl  boingboing.net
Netscape 8      0.1          0.1
Netscape 7      0.3          0.7 
Mozilla         1.3          3.5
Opera 8         2.8           *
Opera 7         2.1           *
IE5/Mac         0.2          0.6
IE4/Win         0.2         <0.1
Safari          0.1         10.6

*boingboing's stats package doesn't subdivide Opera into versions - they're currently showing 2.0% for all versions combined.

All figures are percentages, and before anyone asks where it is: Firefox is not a minority browser as far as either of these sources are concerned. While EWS are showing a few hits (and hence listing the user-agent strings) the number of users is negligible.

So, it's pretty clear that Netscape 8 has not made a big splash. It's lurking down amongst the statistical noise where most sources can't even be bothered to detect it.

During July, Firefox overtook IE5 on TheCounter.com, to become the number 2 browser after IE6. They were the last of the major stats sources (that identify browser versions) to show IE5 with more users than Firefox.

Things are muddled slightly by yet another methodology change, which kicked in mid-month. Based on the monthly total, they've gone back to counting visitors again, rather than hits. Make up your mind, guys.

The main effect of the change seems to have been to give a slight boost to minority browsers, at the expense of IE5/6. It's most visible for Firefox, but Safari and Opera got a boost, too.

I suspect that, if they had made that change earlier, Firefox would have passed IE5 in June, or maybe even May.

There are some indications in the August figures that the July figures might have been slightly messed up by the change in methodology, but it's too early to tell for certain. Treat the July figures with some caution for the moment.

Browser    May   June  July  August*
IE6        83.0  83.9  82.7   83.3
Firefox**   5.4   5.7   7.7    7.6
IE5         7.8   6.5   4.8    4.5
Safari      1.0   1.0   1.5    1.5
Unknown     1.1   1.1   1.2    1.1
Opera       0.5   0.5   0.7    0.7

*August figures are for August 1-10, and should be treated as provisional.

**Due to changes in the browser descriptions shown by this source, there are no Firefox figures available for May. The value shown is for "Mozilla", which is presumed to be Firefox.

A week or two back, w3schools.com decided to change their counting method. They've also re-calculated all their historical figures using the new method. For comparison, here's the old and new versions of their june numbers

browser     old    new
IE6         60.7   65.0
Firefox     26.3   20.7
Others       4.0    2.8
Mozilla      3.5    2.9
IE5          2.8    6.8
Opera 7/8    1.8    1.2
Netscape 7   0.7    0.6

The main changes are obvious - a large reduction in share for Firefox, and gains for IE 5/6.

Shortly after the change, they included a statement on their main stats page describing what they had done. Like a fool, I didn't take a copy, and they've now deleted it. If memory serves, they indicated that they had swapped from counting visits to page views. Their reasons for this are probably related to the disclaimer that appeared at the same time, and remains on the site:

Why so high Firefox figures?

W3Schools is a website for people with an interest for web technologies. These people are more interested in using alternative browsers than the average user.

W3schools have shown anomalously high figures for minority browsers for quite some time, presumably for the demographic reasons I've outlined previously. As a result, many Firefox cheerleaders have taken to quoting their figures frequently, and referring to w3schools as a definitive source.

I suspect that w3schools are getting some flak from people attacking their statistical accuracy (the anti-cheerleaders, if you like - anyone who's seen a usenet flamewar in full swing will recognise both camps), and have added the disclaimer I've quoted above to make their demographic make-up more obvious, and really highlight to casual visitors that w3schools aren't even claiming to be a reliable source of global numbers. Hopefully this will deter some of the ill-informed cheerleaders from telling the world that Firefox has 25% market share.

Apart from that, I suspect that all the referrals from the cheerleaders were leading to loads of Firefox-using visitors passing through that single stats page, without touching the rest of the site. Combined with the rampaging hit-generation machine known as slashdot, it's possible that the visitor stats were being disproportionately affected by the cheerleaders. Swapping to page views would enable their stats to provide figures more representative of the people who are actually using the site (i.e. visiting more than one page). From the persective of w3schools' own web developers, the latter group are far more interesting.

Via libertango, Coyote Blog has an excellent overview of a demographic/statistical phenomenon which is particularly common in the field of browser statistics:

sometimes we would ask our recruits, typically Ivy Leaguers, to come up with the market size for annual snow ski sales. So they would work through the logic that there are 300 million people in the US and x% ski and these people replace their skis on average every 5 years, etc. However, it always made me laugh that these folks would be guaranteed to miss the number way, way high. Why? Because in coming up with the percentage of people that ski, they would look around the room and say, well 80% of my friends ski and so lets assume 30 or 40 or even more percent of Americans snow ski. In fact, I have not looked up the number lately, but the actual percentage of Americans that ski is something less than 5%. Recruits intuition was fooled because, at least in terms of skiing, they were surrounded by an anomalous population.

The 22 May XitiMonitor report (Original French, babelfished English) I pointed to in my last post is notable because it also contains a few figures for other parts of the the world. Now these are for Firefox only, and come with a bunch of caveats (see my previous entry for my thoughts on that subject) but it's still interesting to see comparable stats for Europe and the rest of the world.

Country           Firefox
New Zealand        19.90
Malaysia           15.40
(European average  14.08)
USA                13.98
Saudi Arabia        2.23
Peru                1.92
South Korea         1.73

The figures for New Zealand and the US are reasonable by comparison with Europe; the low figure for Saudi Arabia is probably down to a lack of an Arabic-language version of Firefox, but there are a few surprises in that list.

South Korea is a very tech-literate country (it has a huge number of online gamers, for example) so my naïve expectation would be for a relatively high level of take-up of a new browser. However, that's not what we see - quite the opposite, in fact. Comments from Koreans in the SpreadFirefox.com thread discussing the XitiMonitor report explain a lot:

In Korea, many web sites are IE-dependent. Every shopping mall, banking, online game site uses ActiveX. Also, portal sites that many korean use do not make websites with [...] W3C-based standard tags. Some people may try Firefox once, but they soon recognize it's too hard to use Firefox in the situation like this. So, they delete firefox, and use IE again.

I know from past experience with the Mozilla project that similar factors account for Gecko's support of the (Microsoft-proprietary) <marquee> tag. Quite a few web standards weenies were rather upset when Mozilla started to support marquee. However, reports from those responsible for localising mozilla-based browsers indicated that marquee is used a lot in China, and that without it, mozilla would have serious difficulty gaining any market share at all. Unfortunately for Firefox's Korean hopes, Gecko is much less likely to support ActiveX in any meaningful way for the forseeable future.

Malaysia is another surprise. In terms of population, Malaysian is one of the largest languages not supported by Firefox - yet Firefox still appears to have higher market penetration there than in much of Europe. Again, a comment at SpreadFirefox.com sheds some light on the matter:

Mostly we use the English version. The languages spoken here are basically: English, Malay, Chinese (several dialects), Tamil. So I suspect if not English then we are probably using one of the Chinese localizations.

Peru, like several of its neighbours, has a very strong open-source movement. Those watching the IT press will see occasional reports of this or that country debating legislation requiring government bodies to use open-source software - well, Peru is one of those countries. Again, a naïve interpration leads to an expectation of high Firefox usage; yet that is not the case. Without more information (a full list of browser stats would probably be particularly illuminating) I can't comment further.

Xiti's figures may not be terribly rigorous statistically, but right now these are the only numbers I've seen for several of these countries, and beggars can't be choosers. Interesting, nonetheless.

Following on from their initial study of Firefox usage share in western Europe, French stats provider XitiMonitor have released two more sets of numbers for a wider area of Europe. Again, these are both based on a single (weekend) day's usage, and will therefore reflect personal use more than corporate.

[Sources: 24 April, 22 May, both in French]

Country		24 April     22 May
Finland          30.91       30.05
Germany 	 22.89       24.36
Hungary 	 19.42       22.06
Czech Republic	 18.86       19.93 
Poland           17.92       19.78
Sweden           16.06       16.31
Switzerland      14.32       15.19
Austria          11.28       14.79
France           13.37       14.42
Greece 	          8.96       13.11
United Kingdom   12.12       13.03
*Norway          11.95       12.66
Belgium          10.96       12.16
Romania          11.12       12.01
*Estonia         12.25       11.84
The Netherlands  10.68       10.43
*Ireland         10.02       10.56
Spain            10.01       10.70
Denmark          10.01       10.85
Portugal  	  9.82       11.22
Italy             9.21       10.29
**Luxembourg     14.81        9.64
*Lithuania        3.49        7.00
***Monaco         8.94        5.52

higlighted over 20 million population
*less than 5 million population
**less than 1 million population
***less than 50,000 population

Both sets of figures are based on a sample of approx 23-24m visits, and appear to be an aggregate of all the sites using XitiMonitor's stats analysis service. If, as seems likely, the sample size for an individual country is simply down to the number of visits recorded from that country, then it seems likely that the accuracy of the figures for some countries may be pretty low, due to small sample size.

To pick this out, I've marked large and small countries in the table above. This is based purely on population, so it doesn't take into account a country's "Internet population", which could vary considerably. Most of the large changes (especially the large drops) are in smaller countries, and all the large countries show a similar upward trend. Given that Monaco is over a thousand times smaller than the larger countries on this list, I'm not sure its figures are even statistically valid. The numbers for Luxembourg are probably pretty shaky, too

Again, it's worth noting that the Firefox numbers in this survey are considerably higher than figures from other sources covering the same countries. For example:

Germany, daily figures from webhits.de. These are both Saturdays, rather than Sundays:

              23 April        28 May	
Firefox         9.8           10.7
Other Gecko     6.6            6.4

and Poland, weekly figures (Tuesday-Monday inclusive ending on the date shown) from ranking.pl:

              25 April        30 May		
Firefox         9.8           10.8
Other Gecko     1.8            1.8

The difference from XitiMonitor's figures could be due to the "weekend effect" eliminating corporate use (though less so for the German figures, which are for Saturdays). In theory, corporate IT departments are more likely to enforce homogenous browser installation within their companies, an effect which has (to date) favoured Microsoft Internet Explorer. It could also be down to counting anomalies - both the alternate sources above have good browser detection, and the figures shown are for Firefox only. Adding in the other gecko browsers seen by these sources closes the gap somewhat (and explains the difference between the two countries which isn't apparent in the local stats). I suspect that both these factors are combining, though, and giving anomalously high shares to Firefox.

A couple of conclusions: XitiMonitor appear to be deliberately reporting figures which boost Firefox's share. This is unfortunate, but at least they're up-front enough to point out the weekend effect.

I'm more worried about the browser detection. Even allowing for the weekend effect (which my gut says is probably about 10-20% of the Firefox share), their numbers are still high compared to other sources, and the anomaly looks suspiciously similar to the number of other Gecko users.

That said, even if XitiMonitor's numbers are off by a factor of two in Firefox's favour, usage in Europe is significant, and increasing steadily.

Yeah, I've not posted much here recently. Two reasons: real life intrudes, and there's not been a vast amount of interest to comment on. Browser market shares tend to work on a timescale of months-to-years, so there are always going to be quiet patches. Anyhow, more soon...

http://adtech.info/NewsPresse/Presseintro.php?lang_id=en&node=4

Adtech AG is a German banner-hosting provider. They publish infrequent press releases giving details of stats recorded on their servers.

What They Count

Hits on European banner adverts.

Measuring period and update frequency

Variable periods, infrequent press releases.

Browser Detection

Their browser detection seems moderately competent, but can vary from one release to the next, making long-term comparison different.

Demographics

The major restriction is geographic - their banners are carried on sites based in Europe. This does mean that they are far less English-language-centric than many other sources, though. Like any other banner provider, they will only "see" users who visit sites which carry their banners, which probably means mostly professional (but not corporate) sites.

The main metric they are interested in appears to be the percentage of users of a particular browser that click on their adverts. Given their business, that's hardly surprising, really. Fortunately for us, most of their releases also include a breakdown of browser share.

[About TheCounter.com]

After a brief (by their standards) two-month outage, TheCounter.com are up and running again, and they have back-calculated to fill in the details to when they went down (in early February). Here are their March figures:

Browser     Share
IE6         81.4
IE5          9.2
"mozilla"    4.9
Safari       1.0
Netscape 7   0.8
Opera        0.5
Other        2.2

No real suprises there. However, they seem to have changed their methodology while they were away. At a guess, they've returned to counting hits, rather than visits. Whatever the reason, the difference is easily visible in the "total" count over the past few months:

  Month      Total 
Aug 2004    93656199
Sep 2004    92237275
Oct 2004    93237501
Nov 2004    90515566
Dec 2004    49790829
Jan 2005    61253251
Feb 2005   189085503
Mar 2005   265558858

The disruption in December/January (as discussed here a few months ago) is also obvious. April 2005 looks like it will (roughly) match the March figure, so February is definitely low - I suspect the figures given are only for the last 2½ to 3 weeks of the month.

The March/April figures are in the same ballpark as the numbers for May-July 2004, which implies that they have reversed the methodology change that took effect in July/August 2004.

http://www.artlebedev.com/tools/browsers/

What They Count

Hits on Russian-language banner adverts.

Measuring period and update frequency

Daily, no history.

Browser Detection

Not that good. The headline figures are IE/"Netscape"/Other, with IE broken down by major version, and "Nescape" split by major version with a separate "Mozilla" category.

Demographics

The link to Russian-language content will obviously have an effect. Less obvious is the fact that it will only "see" users who visit sites which carry banners.

The stats can wobble quite a bit from day to day. I think this is mostly a day-of-the-week demographic effect, but there's probably a healthy dose of random noise in there, too. Sample size seems to be ~3m banners/day at the weekends, ~5-6m during the week

[last updated: 2005-04-23]

[This is one of an irregular series of posts detailing sources of free web browser statistics. It is back-dated into the prehistory of the web to allow easy access to the entire series]

French source XiTiMonitor have released some more figures. This time, they concentrate on Firefox, but expand their scope beyond their native France. The figures are for a single day (Sunday March 6th 2005). Note that this will emphasise private use over corporate use; as I noted in my comments on XiTi's last press release, this represents the peak Firefox usage share for any week.

Country  Firefox
Germany	  21.4%
France    12.2%
UK        10.9%
Spain      9.0%
Italy      8.6%

Based on XiTi's last release, the French figures are about 1% above the weekday level in that country. I suspect other nations follow a similar pattern.

I do have some concerns with these figures, though. Here are figures from German stats source WebHits.de from about the same time (week ending 12/03/2005):

Browser       share
IE6           70.1
IE5            8.3
Firefox        8.3
Netscape 7     3.6
Mozilla        2.8
Netscape 4.x   2.2
Safari         1.9
Opera          1.2

Now these are (I think) weekly figures, so I would expect them to be a little lower than the Sunday figures quoted by XiTi; but they're a lot lower. I reckon XiTi are failing to distinguish the individual Mozilla-based browsers, and the figures they're quoting are for the whole Gecko family. From the WebHits figures, all Mozilla browsers (including less-common ones not listed above) come to about 15.3%. The weekend effect could easily add 1-2%. This gets the total up to about 17%. Still well below the 21% in the XiTi figures, but less of a stretch than the headline 8% for Firefox.

There are indications that back up XiTi's German figures, however. For instance, these numbers from Spiegel Online, a German magazine site.

Date          Firefox
15/12/2004     25.3%
01/03/2005     28.3%

[read from this graph]

I'm not sure about whether their browser detection is distinguishing Firefox correctly, but there's no doubting that gecko-based browsers are a serious chunk of market share, as far as they're concerned. Note also that Spiegel Online is a general news magazine, not a tech-geek haven. Germany has a very active open-source community, and several of the more visible battles between Linux and Windows in the last few years have centred on the purchasing decisions of German local government. It looks like that attitude carries over into the browser wars...

[via Tristan Nitot and Pike at SpreadFirefox]

[About ranking.pl]

Another look at stats beyond the English-speaking world.

Of all the stats sources I watch, ranking.pl has been one of the best-behaved; large sample size, regular updates and excellent browser detection. I trust their figures better than most. The figures shown below are a snapshot of their global stats (which comprise roughly 80% Polish visitors and 20% foreign) for individual weeks over the last few months.

Week Ending      18 Oct  15 Nov  13 Dec  10 Jan*  7 Feb*  7 Mar
IE6/Win           76.1    76.3    76.1    77.0    76.0    75.2 
IE5/Win           14.1    12.9    12.3    10.8     9.9     9.3
Firefox            2.3     3.3     4.2     5.1     6.6     8.2
Opera              3.7     3.9     3.9     3.8     4.2     4.2    
Other Mozilla      1.9     1.9     1.8     1.6     1.6     1.6
Everything else    1.9     1.7     1.7     1.7     1.7     1.5

Compared to the US-based stats that have been getting attention recently, there are a few things to note here:

• Opera is doing relatively well (4%, most English-language stats have it below 1%). This is nothing new (it's had a strong showing in these figures for some time) and it's not suffering under competition from Firefox.
• Firefox is soaring (more than trebled in 20 weeks, and no sign of slowing down); it's stealing a few users from other Mozilla browsers, but IE is the main victim.

At the present rate of growth, Firefox will probably overtake IE5 by the end of March, and break the 10% barrier shortly afterwards. Seeing figures like this coming from outside the usual English-language bubble makes me really curious about what's going on elsewhere in the world.

*There's an anomaly in the statistics running from about Christmas though to mid-February. Nothing huge, but it's big enough to swing some figures by about 1% off their long-term trends (for example, the January figure for IE6 is rather high, while the figure for Opera seems a little low) I've not figured out exactly what's going on. At the start of the period, it's probably actually just the demographic effect of Christmas and New Year. Also in this period, a list of names from communist-era secret files escaped onto the web, spurring a frenzy of Internet activity in the country.

However, neither of these effects appears to totally explain things. Every report I can find dates the secret file release to about 28/29 January, yet the biggest distinct "blips" in the stats are in the weeks ending on 10 Jan and 24 Jan. I suspect that the former is a browser-detection change (Opera 8 appears in the figures for the first time at that point) and the latter may be, too.

[About WebSideStory]

For-pay internet stats outfit WebSideStory have released one of their infrequent "data spotlight" columns, and it's causing some minor waves. This is presumably because Firefox is flavour of the month, and this column is titled "Firefox Gains Slowing". [note: this appears to be a static page with no archive; it seems likely that, at some indeterminate point in the future, the content I am quoting from will be replaced by something completely different]

The growth rate of Firefox’s usage share on the web has slowed in recent months [...] Firefox’s market share grew just 15 percent in the five weeks leading up to Feb. 18, 2005, the latest benchmark available. In the previous six weeks before that, the browser grew at 22 percent clip. These figures compare with a 34 percent increase between Nov. 5 and Dec. 3, which coincides with the release of Firefox Version 1.0 on Nov. 9.

Anyone see the mathematical flaw here? The text above is built on an assumption of steady geometric growth. That assumption is highly suspect - in my experience, [Edit:] no browser's market share has ever grown exponentially no browser with significant usage levels has managed sustained exponential growth in market share [end edit] in recent (Internet time) history. Usage of the internet itself may have done so, in the mid to late 1990s, and that might have dragged the browser du jour up at a similar rate, but after 1998? No chance. I speculated on some possible reasons why back in November. So I did some number crunching. Here are their headline raw numbers:

Browser         18 Feb 05  14 Jan 05  3 Dec 04   5 Nov 04
IE 	 	89.85       90.28      91.80      92.89 
Firefox 	 5.69        4.95       4.06       3.03 
"Netscape"*      2.47        2.64       2.83       2.95

[*They've actually labelled the last row "Non-Firefox Netscape and Mozilla browsers". Personally, I think lumping Netscape 4 and Mozilla 1.7 into the same group is fairly braindead, but...]

Unfortunately, the date intervals are irregular - 28, 42 and 35 days. So directly comparing the growth in each interval is pretty bogus in itself (especially with the galumphing great demographic elephant called "Christmas" slap in the middle). Here's the growth numbers I calculated from their data:

Date        Interval    Usage	growth%	 Growth/day	
05 Nov 04     -	         3.03
03 Dec 04     28	 4.06    34.0      0.037
14 Jan 05     42	 4.95    21.9      0.021
18 Feb 05     35	 5.69    15.0      0.021

The "growth%" column is obviously what they're quoting in their article. I calculated it using the formula 100x((usage2/usage1)-1). For that number to be "constant" requires geometric growth in usage from one datapoint to the next.

The "growth/day" column is the average growth per day over the preceding interval - calculated with the formula (usage2-usage1)/interval. I think it shows a much more realistic picture (based on these numbers). Namely: after an initial surge in the first interval, growth in Firefox usage slowed, but then stabilised to a linear increase at a lower level. Since the first interval included the Firefox 1.0 release and a huge publicity push, this seems reasonable.

[Update: some extrapolation]

Back in December 2004, it seemed Firefox was a lock to reach 10 percent by mid-2005, ahead of the reported year end goal of the Mozilla Foundation. Given the latest growth rates, the year end target still appears attainable, but a mid-year achievement is unlikely

Assuming continued constant linear growth at the rate seen in the third interval, Firefox will hit 10% in September 2005. Of course, that assumption is almost worthless...

[Disclaimer: this is all based on a grand total of four data points, compiled using unknown methodology from an unknown data set. Guesswork-a-go-go, in statistical terms]

[28-Feb-2005 update: following a prod from David Naylor, WebSideStory have amended their report to more accurately reflect the numbers. Looks more like a mistake than anything deliberate...]

http://www.websidestory.com/services-solutions/datainsights/spotlight.html

What They Count

Unknown. Like many for-pay sources, they keep details of their methodology under wraps.

Measuring period and update frequency

Irregular, no history.

For those of us who don't pay them, WebSideStory only release the occasional "data spotlight" column. More frequently, their stats are quoted in the press, though these quotes are often partial.

Browser Detection

Seems good, but given the patchy reporting of their figures, it's difficult to tell.

Demographics

Since they don't tell us their methodology, it's difficult to judge demographic effects. They do seem to concentrate on the US, but the data set does appear to be sufficiently large and diverse to avoid any other overt demographic effects.

[About TheCounter.com]

[This post is a follow-up to Next stop, trip-out city and Guessing games]

The counter.com have fixed their browser detection, and their February figures are back in the real world, though they seem to have stopped updating after the first week of the month (hey, if it ain't one thing...). Here are the February figures; I can't really include others for reference, as the changes in methodology mean any direct comparisons would be pretty meaningless:

Browser      02/05
MSIE6         80.8
MSIE5          8.3
Mozilla        5.6
Unknown        1.5
Safari         1.2
Netscape 7     1.1
Opera          0.7

So, how do my guesses hold up?

The first suprise is the Magical Mystery Browser with 5.4%. Given the 5% figure, and the corresponding drops in Mozilla, "Netscape 5" and Netscape 7, I reckon that our nameless one is better known as Firefox. That said, how could one browser be mistaken for three?

Close: the Magical Mystery Browser is actually Mozilla, but it now includes firefox, where it didn't before.

IE1 at 1.2%. Is this a squadron of aviating porcines I see before me? ... My first guess for the real oddities is that "IE1" is mis-identified Safari.

Result! (OK, I wasn't totally certain, but...)

Mozilla's "fall" to 1.2% is almost certainly down to the detection being revised to exclude Firefox. The drop of Netscape 7 all the way to zero seems wrong, though. It's not wildly popular, but it should be above Netscape 6. The October figure of 0.8% seems about right. I think the new browser detection is including it with something else (probably Firefox or Mozilla)

Close, but no cigar. The "mozilla" shown in January actually appears to have been Netscape 7.

There also seems to have been some sort of underlying methodology change, which has boosted the figures for several of the minor browsers a bit. It's most visible in Safari (maybe Netscape 7 too, but given all the detection shenanigans around the gecko browsers, who knows?) I'm not sure what's going on there.