Our company is beginning to do some online direct marketing.
I have good basic knowledge of statistical principles, multiple regression, etc.
We have a landing page ("the creative") that has 10 factors, and 3 or
4 options for each factor.
I can set up a test grid, but I want to test how each option in each
factor might interact with the others without the impossible task of
testing each option in each factor against each other.
Thus, I would like to perform multifactorial analysis. I know there
are statistical techniques where you can select a manageable number of
relevant cells and use those as your test cells, and be able to infer
the likely results for the rest of the grid. But despite poring over
many stat textbooks, I am unable to pinpoint exactly how this
technique works.
I am not looking for someone to actually set up the grid -- just to
point me towards the technique that will allow me to do this.
Any help would be appreciated. Thank you!

---

Request for Question Clarification by mathtalk-ga on 06 May 2004 09:44 PDT

Hi, needstatshelp-ga:

These sorts of problems, which can be labelled "statistical designs"
for the convenience of calling them _something_, can involve difficult
constructions.

A previously posted Question here:

[Q: Calculating n-wise testing combinations]
http://answers.google.com/answers/threadview?id=240204

led me to find a site that offers "a patented algorithm" approach to
such constructions in the form of a licensed Web service:

[Telecordia AETG Web Service]
http://aetgweb.argreenhouse.com/

In the simplest case one might ask for a set of treatments in which
every option is tried at least once (1-wise coverage).  This is of
course trivial to arrange and provides no promise of information about
interactions among factors.

The next case would be pairwise coverage, in which every combination
of two options from any two factors appears (at least once) among the
treatments.  Much more is known of about this problem than about
higher order coverages.

It is possible to write a program to intelligently search for pairwise
coverage designs.  Perhaps for the list price offered you would like a
Researcher to point you to some resources on the Web or in books that
discuss the problem and solution techniques?

However your desire "to perform multifactorial analysis" hints at an
interest in n-wise coverage patterns for n > 2.  Please advise if this
is the case.

regards, mathtalk-ga