I'm going to hazard a guess here, but I think there's only one
certifiable statistics expert on this list. I know who I'm going to trust,
given nothing else to go on. :) Here's my amateur take on the subject...
and my vote to keep analog away from "fuzzy" numbers. As if this was even
being considered seriously.
For any of the "Vector" or "Stickiness" reports to even approximate
accuracy, "User Session" data must be accurate, which is problematic.
First, passing tokens is not 100% effective. Second, referrer data is not
always trustworthy. Third, the analysis tool would have to represent the
structure of the site in some way- which is problematic at best and
impossible at worst. (and no matter what- slow to process)
Even with 100% effective User Session data, you still don't know if I
popped open a window to your site, then switched to my yahoo stock
portfolio for 15 minutes, and then back to your site to follow a couple
links. Or maybe I have my web robot (that reports itself as Netscape)
snarf a few pages for me to read later. One session is overly long, the
other is overly short- and both situations are reasonably common.
Never mind proxy servers. As we all know well, the world's biggest ISP
also runs the world's biggest proxy server system. AOL alone makes most
"Session" data laughably bad, thus dooming "stickiness" numbers.
If you're going to commit to a "fudge factor" then why not simply use
that factor against some concrete metric- a ratio of requests for pages to
unique IP addresses during your site's busiest 15 minutes maybe? I'd say
that number would be at least as valid a measure of "stickiness" as
anything else and is a TON less intensive computationally.
-=Jim=-
P.S. What people want, or even will pay for, and what is actually valuable
are not always the same thing. The judge of value in this case is the
programmer of Analog, and his stance is clear.
------------------------------------------------------------------------
This is the analog-help mailing list. To unsubscribe from this
mailing list, send mail to [EMAIL PROTECTED]
with "unsubscribe" in the main BODY OF THE MESSAGE.
List archived at http://www.mail-archive.com/[email protected]/
------------------------------------------------------------------------
