Hi Stephen,
thanks for thinking about this. here's my argument for it...
>
> > I've been using Analog to do log crunching on some large
> log files and then
> > importing the computer readable format to Access to do
> specialized analysis.
> > It works really nicely and we've been able to generate some
> very helpful
> > results. However, we fall back on WebTrends to get a
> calculation of "user
> > sessions" and "average length of user session" (in
> minutes). I know these
> > numbers may be a little flaky but they are also very useful.
>
> They are worse than "a little flaky", they are totally bogus.
> Let me give
> you an example.
>
> AOL now allocates its customers a different IP number for
> every request to
> your server. So if one person requests your home page and
> that home page
> has 10 graphics on, Webtrends will count it as 11 separate
> visitors. You
> still want to use it knowing that?
Well, i'd always prefer to use analog!
Basically, all log stats are estimates of what we _really_ want to know. We
would really like to know 1) how many different people visit a site, 2) how
many come back, 3) how much stuff they look at, 4) how valuable they find
that stuff to be. for 1) we use a version of "unique addresses" but
understand there are issues with it that move the number in both directions
(the AOL number will increase it, cacheing, and IP-sharing will decrease
it). for 2) we use number of "sessions" divided by number of unique
addresses. for 3) we use page requests and bytes requested (actually i also
like to use a "density" measure equal to bytes requested/page requests).
for 4) value to the user, a good indicator is the amount of time users are
willing to spend on the site. afterall, time is money ;). so we have been
using the webtrends average session length number * number of sessions to
estimate total time and then assign a dollar value to time. I've seen
variations in the web trends number basically ranging from 9 to 13 minutes a
session.
I can't argue convincingly that any of these numbers are absolutely correct,
they all have problems. However, I think the trend is significant so we
track the movement of these indicators month to month and hope to see them
go up.
for requests and "sessions" we'd like to only include sessions that include
a "page request" which might temper the problem of AOL have a session for
each .gif. each page will still show up as a unique session, but that will
be a consistent problem over time. i can even monitor my AOL usage -- if it
is going up a lot, i know it is driving my time/session down while the
number of sessions is going up. my total time (sessions * average length)
should still be okay.
so, i still like the session count and average time numbers. have i
convinced you?
as kind of a p.s., do you know of any resources that help interprete web
statistics? i'd like to know more about things like the general
characteristics of unknown domains, average session lengths, general growth
trends, etc. to compare our sites against. i can imagine in a few years,
college statistics will be teaching specialized courses in Internet traffic
analysis...
dave
-----------------------------------------
Dave Witzel [EMAIL PROTECTED]
Forum One Communications http://www.ForumOne.com
ph: 703/237-8537 fax:703/995-4937
>
> --
> Stephen Turner [EMAIL PROTECTED]
http://www.statslab.cam.ac.uk/~sret1/
Normally: Statistical Laboratory, 16 Mill Lane, Cambridge CB2 1SB, England
Until 12/98: Dept of Math & Stats, 585 King Edward Ave, Ottawa K1N 6N5,
Canada
"Ad infinitum, if not ad nauseam." (Interviewee, BBC Radio 4)
--------------------------------------------------------------------
This is the analog-help mailing list. To unsubscribe from this
mailing list, send mail to [EMAIL PROTECTED]
with "unsubscribe analog-help" in the main BODY OF THE MESSAGE.
--------------------------------------------------------------------
--------------------------------------------------------------------
This is the analog-help mailing list. To unsubscribe from this
mailing list, send mail to [EMAIL PROTECTED]
with "unsubscribe analog-help" in the main BODY OF THE MESSAGE.
--------------------------------------------------------------------
