cornwall - 10:28 am on Nov 22, 2002 (gmt 0)

Brett

Having spent years trying to get to the bottom of what log files really mean, I am not going to let go now till I do :)

Given that we could all (probably) agree that log files are not giving you within 20% to 30% of reality. Perhaps I could try to sumarise what we do know, and ask others to add to the list

1. Comparing monthly log files is relative, rather than absolute. In other words it gives you (broad) trends of traffic looking at the site. If log files are up 10%, then traffic is (probably) up 10%. But they do not give an absolute picture of the real number of visits.
This is presumably true also about the relative importanof referral sites changing over time. If you get twice as many referrals from siteA this month as last month, then in relative terms siteA is delivering twice the traffic

2. Broadly you can ignore browser excentricities other than IE (and to a lesser extent Netscape). Any peculiarities in IE reporting will swamp other browser errors.

3. Bretts point

It's been my experience that 50 to 75% of insite referrals are not correct. Bookmarks, typed-it-ins, drop down history from address bar, caching, no caching, and reloads have turned insite referral numbers to junk. There are no major log file analyzers that have this fact figured out.

4. For a "cachable" site you can be 30% off in you log files
[webmasterworld.com...]
This is a very large figure

5.

no cache mania. Most of the dsl, cable, and other high speed modem manufactuers are telling people to turn off caching in their browser. They all have explicit details on their site as one of the setup steps to take. That in turn is skewing referral numbers as even a simple back button can cause a page reload. That referrer will often be the previous page.

It would be a pity to think that log file analysis is rubbish, but it is difficult to draw any other conclusion that we do not really have any sound idea where web traffic is really coming from.

But it perhaps goes towards explaining why Webtrends and others have not been keen to answer the question!