[Sigia-l] Less than one-minute visits

[Beth Mazur]

> Reviewing the reports of a new client's website, I found that for one week
> almost 75% of visitors were on the site for less than one minute! 

Others here have offered some explanations for some of this. I cannot pass
up the chance to point that the noise in this stat (particularly for the old version 
of WebTrends that churns log files) is probably so high as to warrant a
huge grain of salt when you consider it. In the live version of WebTrends 
(which uses web bugs), the average visit length is total time spent on pages 
divided by total visits. Assuming that the older version does something
similar, I'd be really cautious, as its method for computing visits is
particularly problematic.

In other words, garbage in, garbage out.

Analytics fans have argued that even though this stat measures all sorts
of junk (like how much time people are IMing their friends, going to the 
loo, and/or watching American Idol while surfing), any site with a decent
volume will have these noise effects muted and that in any case, the real
use is to view the trend. Maybe.

And of course, having a long average visit length isn't necessarily good. 
How do you know whether a long visit is due to site interest *or* having 
it take ten minutes to find something?

So, if I can channel my usability friends, the real answer is to find out from 
representative visitors. 

In your case, you may be able to find out that it is some simple explanation
like folks are actually being successful or you're not filtering out the internal
users (which as someone else said, you can do in the old WebTrends).
Then again, you may well have a problem with your site. (What's
the home page reject rate?) We had a problem once where a stupid
JavaScript bug showed up only for AOL users (a chunk of our audience).
Needless to say, having a JS error pop-up on the home page, given
folks concern about security and privacy, was not a way to entice folks
to keep browsing. Oops!

Beth Mazur
IDblog: http://idblog.org