Hacker News Reader

Comment by oneeyedpigeon | Hacker News Reader

NetMarketShare's stats wildly contradict everyone else's [1]. If you believe them, IE is used almost 3 times as much as Chrome/Firefox. If you believe anyone else, Chrome is ahead of IE.

[1] http://en.wikipedia.org/wiki/Usage_share_of_web_browsers#Sum...

View on HN

One of the reasons that IE marketshare is so hard to measure is that it depends on whether business users visit the site. You can see this if you can get data that isn't filtered. The only public data I know like that is Clicky [1]. Note the sawtooth wave caused by users who use IE7/8 (and now IE9) during the week but some other browser during the weekend. So IE8 marketshare is almost 2x higher during the week than on weekends.

http://clicky.com/marketshare/global/web-browsers/internet-e...

The same wikipedia page gives the methodology applied for those calculations.

- StatsCounter (hits, not unique visitors)

- W3Counter (last 15,000 views from 50,000 websites)

- Net Applications (stats from 40k websites having 160 million unique visitors per month)

- Wikipedia (page requests)

Only NetApplications is using unique visitor stats, the rest are tracking views. The results are in line with what most people would assume; Chrome and Firefox users are more active users. But in terms of number of unique visitors IE is still far ahead.

> The results are in line with what most people would assume; Chrome and Firefox users are more active users.

It's not true. top.mail.ru (large Russian counter, ~250M views per day) publishes both number of visitors and number of views. Visits/visitor ratio are about the same for different browser families. Also, stats from top.mail.ru are very similar to the stats from StatCounter on Russian Federation [2], while Net Applications hide their per country totals behind pay wall, and their China-weighted totals are completely outlandish.

[1] http://top.mail.ru/browsers?id=250&period=0&date=&aggregatio...

[2] http://gs.statcounter.com/#browser-RU-monthly-201210-201310

The same wikipedia page has a lengthy discussion on the questionable nature of NetApplication's stats: http://en.wikipedia.org/wiki/Talk:Usage_share_of_web_browser...

NetMarketShare's stats are more accurate than any of the others listed there. The mere fact that W3Counter, StatCounter, and Wikimedia are listed there shows that whoever compiled it has no understanding of global web stats. Chrome is only the dominant browser in two countries and elsewhere only dominant amongst a specific subset of demographics.

If that's true, maybe you could correct the Wikipedia article? Honestly, I'll hold my hands up: I don't have an in-depth understanding of how each of those stats companies gathers their data and, having heard widely-reported statements about Chrome exceeding IE's market share, my natural instinct was to accept the majority's figures over the minority. If I - and, no doubt, many others - are being misled, it would be nice to correct our misconceptions.

I gave up editing Wikipedia years ago after someone kept subtly vandalizing a page for one of my big websites and removing mentions of it on other pages which had been placed organically by multiple editors in favor of a competitor's site in another country operating illegally. To the point that demonstrably wrong information was in wikipedia and demonstrably correct information was removed. This went on for months and there wasn't a thing I could do about it, so I just stopped bothering. Essentially, most pages wind up self-selecting among people with similar viewpoints or people who feel like sticking it out the longest in edit wars.

As far as the wikipedia stats links, jeswin correctly points out why the other measures are inferior (and useless when talking about overall web usage) above. You're free to update the page to reflect this, but I think I've had enough of wikipedia egos and edit wars to last a lifetime.

They are not "more accurate". Just different. They measure 2 different things. I think pretty much any website owner can tell you their stats look a lot closer to what StatCounter says than what Net Apps says, though.

> NetMarketShare's stats are more accurate than any of the others listed there

On every site I've monitored (none of which cater to technical users, one of which is aggressively non-US-centric), NetMarketShare's stats wildly inflate IE's market-share, particularly internationally. That's not to say that they're fictitious but rather that they reflect a certain viewpoint and methodology which is only relevant for a few customers. I've found Akamai's numbers to track very closely with what I measure:

http://www.akamai.com/html/io/io_dataset.html

They don't publish technical details about their methodology but I suspect you're seeing a sampling bias: NetMarketShare is used by a modest number of large companies, almost exclusively North American and mostly related to media, and their stats appear to depend on an embedded JavaScript tracker:

http://netmarketshare.com/faq.aspx#Methodology

There are some apparent technical botches: they didn't appear to use a CDN at all — when I last traced counter.hitslink.com in the spring everything went to Los Angeles and was correspondingly slower globally but it now[1] appears that they might be using CloudFront — or set Cache-Control headers, so if you're outside North America or on a slow connection you were going to see a higher rate of their JS failing to load and for the tracker ping back not to load by the time the user navigates to a new page:

http://redbot.org/?uri=http%3A%2F%2Fcounter.hitslink.com%2Ft...

(In contrast, the Akamai numbers reflect direct HTTP traffic for sites like microsoft.com, iTunes, etc. so they avoid the issues related to JS trackers but do require care to filter robots which spoof a browser UA)

They also listed some major sites like CNN or Mozilla which didn't use that JS bug so they either have an unadvertised bulk log delivery service or are portraying smaller side sites as representative of the massive high-traffic main sites.

The real question is the way they interpret the data, and in particular the weighting scheme. Because “everyone knows” Internet Explorer is common in China, they weight the numbers to represent this but you're simply hoping their usage of the CIA Internet Traffic by Country report is accurate:

http://www.netmarketshare.com/faq.aspx#Country

The numbers I can measure are very close to Akamai, showing Chrome tied with IE even in China:

http://www.akamai.com/html/io/io_dataset_v2.html#stat=browse...

1. http://www.webpagetest.org/result/131107_2C_PZK/1/details/

Akamai's numbers are simple % of requests, which means 'hits'. So, a single Chrome user visiting 10 pages that happen to make 100 CDN requests is counted 10x as much as an IE user that visits 2 pages that happen to make 10 CDN requests. In short, the numbers are useless if you're looking for % of visitors.

NetMarketShare uses visitors, which is preferable if you're looking at percentage of users, which is what we're discussing here.

Your point is currently technically correct for the public dataset but it's not significant unless the two browser populations have highly variable usage patterns, which is only going to be true for something like a Chrome fan site. As an example, Google Analytics shows visits, unique visitors or page-views all placing Chrome at 85% relative to IE in China within 0.5%.

Again, I'm not saying that NetMarketShare is lying – simply that it's critical that you make decisions based on data which is relevant to your particular site and that you have multiple data sources to compare and make sure that what you're seeing is a real trend and not an artifact of your analysis.

Reply on news.ycombinator.com