I'm giving up comparing websites using Alexa.com's data. Checking the accuracy of Alexa data using ABCe numbers led me to believe they were good enough to rely on. But the Times reckons the Alexa numbers in my latest post relating to referrals from the BBC were rubbish - and the Guardian agrees.

Alexa data vs the Times's own

Drew Broomhall's tweet about Alexa's data

Alexa had claimed that 4.1% of Times visits are referrals from the BBC.

But Drew Broomhall, the search editor of Times Online, tweeted to say the figure was in fact just 1.4% - just a third of what Alexa is claiming.

Alexa data vs the Guardian's own

Martin Belam's tweet about Alexa's data

Martin Belam, Information Architect at the Guardian, saw the exchange and offered to reveal the Guardian's figure for BBC referrals if I found out the Alexa one.

So I did: Alexa claims that 4.44% of visits to the Guardian were preceded by a visit to bbc.co.uk, making it the Guardian's 3rd most important referring site.

Martin's reply: "Well for a start, the BBC only just scrapes into our top ten referrers for last month, not third highest"

Conclusion

There's only one obvious one - let's not rely on Alexa data for making comparisons of websites ...

Instead of blaming the problems of online news sites on the BBC "dumping free, state-sponsored news on the market", News Corp chief James Murdoch should thank the BBC for all the traffic it sends his way.

The BBC sends about 870,000 visitors a month to the Times website and 1.1 million to The Sun's, according to Alexa.com. It sends similar numbers to Sky's websites.

Percentage of traffic to News Corp's sites from the BBC

Alexa measures visitors' 'preceding sites' - the site they were on before they visit a particular site, and which is likely to have sent them there via a link.

The Alexa figures show how important the BBC is in sending traffic to the Times, Sun etc.

• Times Online: 4.1% of visits were preceded by bbc.co.uk - making the BBC the 3rd most important preceding site after google.co.uk (12.9%) and google.com (12.2%). (See the Alexa data - click clickstream).
• The Sun: 4.35% of visits are preceded by the BBC's site. This makes the BBC the 4th most important preceding site - after google.co.uk (9.5%), google.com (9.2%) and facebook (5.3%).(See Alexa data.)
• Sky.com: 4.0% of visits are preceded by bbc.co.uk, making the BBC the 5th biggest preceding site - after google.co.uk (14.9%), facebook.com (7.8%), google.com (7.7%) and yahoo.com (4.6%). (See Alexa data).
• Skysports.com: 6.6% of visits are preceded by bbc.co.uk, making the BBC the 3rd most important preceding site - after google.co.uk (10.0%) and facebook.com (9.0%). (See Alexa data).

(Reading by RSS? Click here to see the video version.)

Methodology

The figures are based on Alexa's clickstream data, which does have a slight American bias as my post on the accuracy of Alexa data found.

The Alexa data measures preceding sites - the order in which sites are visited. It doesn't measure how users got from one site to another. This is most likely to be because a user clicked a link - in which case the BBC is benefiting the next site to be visited. It could also be because they, say, clicked an existing bookmark, in which case the BBC is not responsible for that visit.

The Alexa figures are for visits, whereas the ABCe numbers are for visitors. However, I multiplyied the Alexa %s by the ABCe numbers to come up with the actual visitor numbers for the Times and Sun.

Alexa.com lets you compare stats about websites. But how representative is its data? It's hard to know as it gives figures as %s rather than absolute numbers.

So, to find out, I've compared Alexa with the ABCe official audited data for UK newspaper sites - using the figure for the %age of each site's visitors from the UK.

As the table shows, Alexa is good but not brilliant.

Alexa vs ABCe

In particular, Alexa consistently underestimates the proportion of users who are from the UK (maybe reflecting its American roots?). However, the Mirror apart, the spread of errors is reasonably consistent.

Using the table:

• Clicking the first column takes you to the relevant Alexa page.
• The Alexa column shows the percentage of users Alexa thinks come from the UK. They don't give the timeframe.
• For the ABCe column, I dividing the June ABCe figures for total unique UK users into the overall total unique users. The FT isn't shown as it isn't audited every month (and wasn't in June). The Express's site isn't audited. The ABCe do some aggregating of data - for instance, the Sun figure includes the News of the World site. Alexa treats this separately.
• The average is a simple average - it isn't weighted in any way.

Using Alexa data

I'm going to use Alexa to make more comparisons about UK newspapers over the next few days.

You can read more about interpreting Alexa data on Alexa's own explanatory page - in particular note that they no longer just rely on people who have downloaded their toolbar, as they did in the past.

Some other information about the data (but much of it written when they did still rely on ONLY the toolbar data) includes:

• Impact of Museum Websites - a comparison using Alexa (see the comments)
• Please stop quoting Alexa data (although see the last comment where the author says things have improved).
• Alexa toolbar and the problems of experiment design.
• And Mark Pack's recent post (March 2010) on how the internet is changing British politics (well worth a read).