Robert Scoble posts an interesting post (which I found via John Naughton, though I subscribe to Scoble too) about his perceived inaccuracy of comScore, Alexa, Compete, etc. I’ve long learned, through subscribing to Scoble’s link blog, that if you mention him he’ll generally find and share your post. So, hello Scoble, and here’s a little bit of info you might enjoy.

Media UK does over 2m pageviews a month, so it’s a pretty good website to check that website reports do the right thing. Since they’re my figures, I can share them with impunity. So, just so you can see how comparable these services are, here are the output of three graphs (showing the last 12 months), which I’ve cropped and resized to enable at-a-glance comparison…

Google Analytics, showing total visits per day.

Alexa, showing daily unique visits per day (not quite the same)

Compete, showing total visits per month, ending in March.

[If you have full Hitwise access, you should hopefully see a similar graph for the last 12 months to GA. If you do, contact me, since I'd be interested to see the graph for 'visits', and publish it here.]

Well.

Given that Google Analytics (“GA”) gets a near-100% coverage of all visits to mediauk.com – the code’s on my site – it should be telling “the truth”.

The first thing you’ll spot from the above is that Alexa and GA both show two major ‘traffic hits’. One is over Christmas, as you’d expect. The other is interesting: I can’t adequately explain why Alexa showed a major reduction in August. According to GA, I had precisely no visits for three days: I removed the GA code by mistake from the pages. So why does Alexa show a reduction too? Do they have some kind of link with GA? Removal of the code didn’t cause any errors causing pages to not complete loading, so that’s out of the frame. Curious. How else would you explain it? Curious.

The second thing you’ll spot from the above is that GA shows a modest increase, whereas Alexa shows a modest decrease. However these figures are derived, it’s my contention that all services should show a similar trend: if total visits increases, it will do on all services. This clearly isn’t happening. Now, Alexa only shows a ‘share’, and it’s possible that total web traffic has increased faster than my own: do we buy that as an argument?

The next thing you’ll spot from the above is that Compete shows absolutely no correlation with the figures you see. The disaster around Christmas (according to GA) seems to result in tremendous figures in Compete.

Compete and GA both return a “monthly total visits” figure. I’ve always believed this is the most useful figure for a website, since it shows how often people are consuming your content, as well as how many people also come to your site (the radio equivalent would, I guess, be ‘share’). It also ignores any page-view issues caused by Ajax and other technologies, and the fact that you’ve hundreds of pages doesn’t skew the figures too: it’s a great metric to use to show website popularity. A ‘visit’, just to be clear, can be one person visiting many times, as well as one person visiting occasionally. Compete and GA both publish visit figures for my website. And the figures are, for last month:
- Compete: 5,000 visits a month
- GA: 439,000 visits a month
That is a not unsubstantial difference. But Compete is, apparently, only measuring US traffic; so perhaps I should only look at mediauk.com’s US traffic, to be fair. So…
- Compete: 5,000 visits from US in March 2007
- GA: 17,568 visits from US in March 2007
So, on this evidence, Compete is under-representing my traffic by over two-thirds – as well as demonstrably not following any trends in mediauk.com’s site traffic. The figures are almost entirely unrelated to my website’s traffic. This is bad. Are websites basing purchase decisions on Compete’s data? In which case, do I have a legal case against them?

We’ve not, yet, mentioned comScore. That’s a blog post for another day, I suspect: because there’s so much more there than meets the eye, it’s not funny.

Tags: alexa, compete, comscore, google, google_analytics, inaccurate, robert_scoble, traffic | Permalink | Comments RSS