How to
Set Up and Run a Web Experiment - Part 2
By
Charles
Hofacker, Professor of Marketing, College of Business,
and
Jamie
Murphy, a student at the College of Communications,
Florida
State University.
Hofacker and Murphy
frequently publish research and commentary on
internet marketing in scholarly journals and the
New York Times Online.
They operate New South Network
Services,
an ISP and Web Creative Firm in Tallahassee, FL
We have been running experiments on the world wide
web since the Fall of 1996, and it never ceases to
amaze us how unpredictable surfers are. We have found
experimental research on web pages endlessly enlightening.
You never know "what makes people click."
In our last column
Who Knows
What Pages Work Best?, we made a case for defining
your site's goals, and then finding out empirically
which site design best achieves these goals. The direct
marketing industry has been doing this for years,
constantly fine tuning their message and tinkering
with their format.
We have initiated
research in the area of optimal page design on behalf
of a number of our clients at New
South Network Services using a simple methodology
that could be adopted by other sites.
Most sites have
a page called "index.html" which functions as the
main entry point, or home, for a web site. When a
visitors asks for http://www.x.com, the server looks
about on the main document directory of www.x.com
for "index.html" and sends this file to the visitor.
We replace this page with a Perl script named "index.cgi."
The script contains
a list of two or more files which represent alternative
versions of index.html. For example, one version may
be white on black and a second version might be black
on white. The script creates a random number, and
picks one of the files based on this number. It then
opens the chosen file and writes it out. That file
is the opening page the viewer sees.
In addition, before
finishing up, the script logs the visitor's domain
name, IP address, time and date of access, and the
file which was randomly chosen. That way it is possible
to determine which version of the home page each visitor
saw. The log file can then be mined for differences
in behavior between the random visitors to the several
versions of index.html in terms of links visited or
other desirable goals.
Let the winner
be proclaimed the official index.html!
