StatCounter: Microsoft's post about Net Applications is "one-sided"

On Sunday, Microsoft posted up a new entry on its official Internet Explorer blog, explaining why the company prefers the methods used by Net Applications to record web browser market share over a rival service, StatsCounter. In summary, the article, written by Microsoft's Roger Capriotti, said Net Applications discounts the visits generated by Chrome's pre-rendering of sites, along with its endorsement of "geoweighting” for handling net traffic and Net Applications using unique visitors over page views.

We decided to contact StatCounter for comment and received a lengthy response from Jenni Cullen, the company's chief financial officer. Cullen told us that a few years ago, StatCounter decided to publish its global web browser stats for free to make how it presents its data "as transparent as possible."

Cullen told us:

No stats service (even our own) is perfect and it’s up to users to decide which service suits their needs. That’s why we try to make our service as transparent as we can and to that end it would be helpful if people provided a balanced comparison between the various services. Unfortunately, in this case the Microsoft article appears to be somewhat one-sided.

Cullen also told us that the company has had "extensive discussions and correspondence" with Microsoft's Internet Explorer team in the recent past and StatCounter was surprised that Microsoft's IE blog article was posted "without any request for comment from us."

Cullen broke down all three arguments Microsoft gave in its blog post about preferring Net Applications over StatCounter. On the subject of Chrome pre-rendering web sites being counted in StatCounter's results, she states:

From our perspective we are, of course, aware of pre-rendering but we haven’t felt that it’s something we need to adjust for at the moment. If you look at the trajectory of Chrome in our stats, pre-rendering introduced in June 2011 (per Roger’s article) did not have any significant impact on our stats at all. There is no bump or sudden jump in Chrome.

Cullen says that at the moment the company has not received any requests from StatCounter users in regards to pre-rendering but added that they are open to feedback from Microsoft and others concerning this issue.

There's also the topic of Net Applications' "geoweighting” method of tracking web traffic and adding in the Internet users of specific nations around the world via CIA data. Cullen points out that Capriotti says in the original artcle, "You could of course use other data sources available to you to account for real world internet population." She states:

Our point exactly! We don’t impose any weighting system on our users – they are free to take our data and weight it however they wish or not at all. The decision is theirs.

The final point in the Microsoft IE blog post was about Net Applications using unique visitors for web browser stats versus StatCounter's page views. Cullen says the company deliberately decided to use page views for web browser stats, claiming there are a lot of problems with tracking unique visitors:

Do you define a unique visitor based on cookies? How do you handle browsers which don't allow cookies? In this case, each pageload is counted as a unique. Or do you use IP address? How do you handle dynamic IPs (such as AOL) where the IP changes with every pageload? In this case, each pageload is again counted as a unique.

Cullen also makes a number of other points about StatCounter versus Net Applications. For one, StatCounter claims it gets its data based on a pool of three million web sites compared with just 40,000 for Net Applications. Also, Cullen says StatCounter tracks both javascript and non-javascript browsers while "to the best of our knowledge" Net Applications excludes non-javascript browsers from its data. Finally, StatCounter counts third party browsers Maxthon and Lunascape as separate programs while Net Applications bundled both in their IE data. Cullen states:

I’m unsure why Net Apps feel it’s meaningful to combine these unrelated browsers. (The stats are regarding browsers – not rendering engines.)

Report a problem with article
Previous Story

Firefox Windows 8 will embrace Metro's framework

Next Story

Rumor: Microsoft bans buying iPads, Macs with company money?

14 Comments

Nice read, but I would take Jenni Cullen's words as - I want to protect my brand. Besides, when it comes to technical questions, always defer to an actual technical person not a public relation/financial mouth piece.

ahinson said,
Nice read, but I would take Jenni Cullen's words as - I want to protect my brand. Besides, when it comes to technical questions, always defer to an actual technical person not a public relation/financial mouth piece.

Sound like very valid points to me. I know from experience that tracking unique visitors is next to impossible. I do get why pre-rendering might be an issue if you want accurate statistics though as it isn't an actual visit.

Geo-weighting makes no sense to me though... what's wrong with just counting the number of users unless you are doing a direct country v country comparison?

Fourjays said,

Geo-weighting makes no sense to me though... what's wrong with just counting the number of users unless you are doing a direct country v country comparison?

Lets say China's population uses 50% IE, 25% chrome, and 25% FF. Meanwhile, Romania's population uses 50% FF, 25% IE, 25% Chrome. You can average that out, and see FF gets 37%, Chrome gets 25%, and IE gets 37%.

But what if your numbers from China report 10,000 people, and your numbers from Romania also report 10,000 people? Meanwhile, you have another stat that says there are 10* the number of internet users in China than Romania? The stats for China should be weighted 10x as important as the stats from Romania, to get an accurate picture of the browser stats.

Fourjays said,

Geo-weighting makes no sense to me though... what's wrong with just counting the number of users unless you are doing a direct country v country comparison?

Because it depends on the demographics of the websites they are monitoring. For example, if a large country like China used Internet Explorer 100%, and has lots of users, but only a couple Chinese people actually visit a site you monitor, you won't be representing the browser marketshare of China at all. It will only count for a couple of votes, even though the internet accessing population of China is huge.

It's not always this one-sided, but geo-weighting would solve this problem. It's a reasonable way to normalize the data.

rfirth said,

Because it depends on the demographics of the websites they are monitoring. For example, if a large country like China used Internet Explorer 100%, and has lots of users, but only a couple Chinese people actually visit a site you monitor, you won't be representing the browser marketshare of China at all. It will only count for a couple of votes, even though the internet accessing population of China is huge.

It's not always this one-sided, but geo-weighting would solve this problem. It's a reasonable way to normalize the data.

You wont really magically represent it by multiplying your numbers either.

It would be like making a survey in United State and giving this survey to 5000 NY city citizen and 10 other people across the country then multiplying the number to reprensent the whole USA. 10 people is not enough multiplying it wont change this fact.

You need to consider the margin of error. You need to check if the sample is good too. Let's say China represent 10% of internet while only 4% of your visitors are from china. Then you need to be sure this 4% is representative (ie is made of people from all around the country and from different social status). If this 4% is from companies based in Beijing mostly because your sites are of an interest to chinese companies only then your numbers might not be good with or without Geo-weighting.

I don't have any problem with Geo-weighting. But i have a problem with MS saying Geo-weighting is awesome without taking the time to talk about the quality of Net applicaiton data.

Edited by LaP, Mar 20 2012, 10:32pm :

rfirth said,

Because it depends on the demographics of the websites they are monitoring. For example, if a large country like China used Internet Explorer 100%, and has lots of users, but only a couple Chinese people actually visit a site you monitor, you won't be representing the browser marketshare of China at all. It will only count for a couple of votes, even though the internet accessing population of China is huge.

It's not always this one-sided, but geo-weighting would solve this problem. It's a reasonable way to normalize the data.


Right, got it now. Makes sense. Thanks

Enron said,
Sounds like neither Statcounter or Net Applications is doing a very good job of what they say they do.

Atleast Net Application does a better job than Statacounter. (Microsoft, 2011)

s3n4te said,

Atleast Net Application does a better job than Statacounter. (Microsoft, 2011)


Net Application is might be better, but just for a few countries. It does not count statistics for more than 9/10 of the world. StatCounter does. But doing it not as good as should. For example Russian and Chinese stats being collected just from few western oriented people, who are not average. It's very one-sided representation. And it looks like they do not add it to world statistics proportionally to country's internet population.

Commenting is disabled on this article.