We use the traffic ranking tool Alexa to keep an eye on what’s happening on the web; it’s a good measure of the speed with which social media is transforming web habits and economics. But most of its users know that Alexa’s a relative measure, so if a chart’s showing downward movement it doesn’t necessarily mean that traffic is falling.

Alexa measures traffic by encouraging people to download its tracking tool to their browser, and then monitors their usage. Most Alexa users know that it indicates a site’s share of reach, rank and page views by revealing what proportion of the Alexa sample has visited the site, and that there’s no clear indication of absolute numbers. Also, Alexa clients tend to be skewed towards techies and other early adopters so it doesn’t necessarily give an accurate view of market share across internet users as a whole.

To test Alexa accuracy and find out what its results might indicate in absolute terms, I decided to do a quick and dirty experiment with a sample of one.

I took a fairly big site for which I know absolute results (no names) and plotted it against the Alexa graph. Here’s the result for the last six months:
I could match the charts for the first three months or the last three months but not for both. There seems to be a step change between them.

This is the first chart, showing a nice correlation for the first three months…

But in November the correlation breaks down. In August, September and October last year 0.1% of Alexa daily page views equated to approximately 7.5m Internet page views, but in November, December and January there was a big shift upwards: 0.1% of Alexa daily views equated to 11m Internet page views. You can see that by shifting the absolute chart downwards I managed to restore the correlation - but I had to move it a lot.

In August, September and October you needed 50k internet page views to obtain 0.0006% of Alexa daily page, but in November, December, January it took 64k page views to get the same Alexa score.

The implication is that sometime between the two quarters internet usage grew by 28%! I don’t believe that, but there’s some kind of big increase going on (called Facebook-bebo-orkut?). It is at least reassuring that the two graphs match quite closely overall, showing that Alexa tracks trends closely, and that it’s taking ever more page views to maintain market share.

Here’s my estimate of how Alexa numbers translate into total (public) internet page impressions. Any views on how bonkers these might be?

[Estimate of absolute value of Alexa daily page views November 2007]