Understanding Browser Usage Share Data

Every month the industry has a standing news cycle that reports on the latest browser market share numbers and each month we provide an update on how we are doing against our goal: the growth of IE9 on Windows 7. As we prepare these updates, we and others in the industry often see trends in the data that are worth exploring. This has been the case over the past few months as the effect of geoweighting and other factors, such as prerendering, on usage data has become increasingly clear.

There are currently two primary sources for share data: Net Applications and StatCounter. While both provide detailed browser usage data, the two are very different when it comes to methodologies and results. Often they show the same general trends, but the overall share numbers they provide for each browser are significantly different. Members of the technology community have begun to pick up on this as they’ve discussed share over the last year, often raising the question of which data source is more reliable. For example, some industry experts have begun to question how much share Chrome really has in the market based on discrepancies in the data.

Looking at the differences in how monitoring firms measure market share can help explain these differences in share and also underscore why we look to Net Applications to track how customers are using our products.

Two different methodologies to measure browser share

1. Real usage versus prerendered non-usage share. Starting in June of last year with Chrome v13, Chrome began “prerendering” certain web pages in Chrome. With prerendering, Chrome is opening separate Chrome tabs based on user search queries at Google.com or from Chrome’s Omnibox that are invisible to the user. If the user then clicks these search links, then the tab and page will display. However, a certain portion of these links will never be clicked and the user will never see them – remaining invisible to them and therefore not real user page views of those prerendered sites. Last month, Net Applications began removing Chrome prerendered browsing traffic from its statistics, noting that “prerendering in February 2012 accounted for 4.3% of Chrome’s daily unique visitors.” In doing so Net Applications became the first company to adjust its data reports for websites Chrome users never visited. At least one writer noted “analytics companies that don’t take into account pre-rendering may be inflating Chrome numbers.” One such company is StatCounter, who simply publishes their data as they record it, without any adjustment for prerendering. NOTE:  StatCounter recently announced that they have updated their data as of May 1, 2012 to reflect prerendering in Chrome.  However, there is no indication of either methodology or what percentage of Chrome share is being removed from StatCounter data.

2. Geoweighting of browser usage based on real world internet populations. Most browser analytics companies that report browser usage do so based on a network of partner sites that help provide them that data, but only one – Net Applications – does “geoweighting” of that data. As Net Applications explains:

The Net Market Share data is weighted by country. We compare our traffic to the CIA Internet Traffic by Country table, and weight our data accordingly. For example, if our global data shows that Brazil represents 2% of our traffic, and the CIA table shows Brazil to represent 4% of global Internet traffic, we will count each unique visitor from Brazil twice. This is done to balance out our global data. All regions have differing markets, and if our traffic were concentrated in one or more regions, our global data would be inappropriately affected by those regions. Country level weighting removes any bias by region.

This is absolutely critical to us in understanding what IE share is worldwide so we can better serve our customers. StatCounter, on the other hand, does not do geo or country level weighting. They simply report absolute global page views. As StatCounter notes:

We do not manipulate the data in any way. We do not collate it with any other information sources. No artificial weightings are used. We simply publish the data as we record it.

This factors heavily into the disparate numbers from Net Applications and StatCounter, since StatCounter’s data is entirely beholden to the coverage of its sites. To better understand what this means, we ran some models.

Net Applications does its country weighting based on public CIA data of worldwide internet users by country. This was an adjustment they started some years ago in an effort to build a more accurate view of the world’s browser traffic, OS usage, search usage, etc. Below is the current CIA weighting for the top 15 countries in the world.

image

Source: The World Factbook, Central Intelligence Agency.

By reporting just traffic, StatCounter instead relies on the location and web traffic of its partner sites versus affording any weight to how many people are actually on the internet in that country. StatCounter recently released that sample size by country for the month of January 2012 for total web page hits.

image

Sources:  The World Factbook, Central Intelligence Agency; StatCounter, January 2012.

You’ll notice some pretty big differences in the weighting of StatCounter versus Net Applications. First and foremost, the most populous country in the world, China, doesn’t make the top 20 for StatCounter, when in fact it represents the world’s largest internet population. Japan, a country that represents the number three country in the world in terms of internet users, barely makes the top 30.

How does this happen? StatCounter only uses total pages views by country, and that number is only as good as the site coverage they have in any given country. That explains why Turkey, a country that is 15th in the world in terms of internet population, is the number two country in the world based on StatCounter’s data sources.

image

Sources: The World Factbook, Central Intelligence Agency; StatCounter, January 2012.

So how does this actually affect StatCounter’s reporting of worldwide browser share? We took a look at StatCounter’s publicly reported worldwide share for February. This is a February report on share and represents a measurement model where not even 1 in 20 Chinese users are counted in determining worldwide browser share. Or where worldwide share is based on a ranking where the United States represents a disproportionate number internet browser share – 24% versus the 13% of the world’s internet population it truly represents.

To further explore this problem, we re-ran the StatCounter numbers and weighted their publicly reported individual country browser share numbers by the CIA internet population data. This calculation would then represent a true country or geo-weighted view of worldwide browser data based on the actual world’s internet population.

imageimage

image

Sources: The World Factbook, Central Intelligence Agency; StatCounter, February 2012.

The difference in reporting of share is stark. This geoweighted view shows that Internet Explorer is under-reported by 9% when real, actual internet populations are not considered. Chrome is over-reported by almost 6%. Firefox is over-reported by 3%.

We used the public CIA country data for this analysis. You could of course use other data sources available to you to account for real world internet population, but however you look at this, the discrepancy in numbers is substantial and helps to explain how two different data providers – Net Applications and StatCounter – can present two very different views of the world’s browser share.

3. Unique visitors versus absolute page views. One last difference between Net Applications and StatCounter is that while StatCounter only reports absolute page hits without any filtering, Net Applications actually reports usage share based on unique visitors. It is this type of analysis that allows them to achieve more accurate representations of browsing habits and actual usage by removing Chrome prerendering traffic (4.3% last month) in an effort to separate real page views from invisible page views. It is also seen as a more accurate way to determine actual browser usage because it is less susceptible to fraud. Wikipedia notes “Measuring browser usage in the number of requests (page hits) made by each user agent can be misleading.” It can lead to overestimation and even fraud in the case of bots that will register large numbers of web page hits. More information is available here at Wikipedia.

Getting closer to the truth on browser share

Certainly there is no perfect way to measure browser share at a worldwide level. This is why we focus on utilizing the data that provides the most accurate representation of actual usage share around the world. While both Net Applications and StatCounter often show similar trends, it’s important to note that certain steps must be taken to represent accurate quantitative assessments of worldwide usage share.

Roger Capriotti
Director, Internet Explorer Product Marketing