David,

this sounds suspiciously like "conjoint analysis," which was documented
pretty well by Paul E. Green, Harvard Business Review, July-August 1975,
pp. 107-117.  And a number of other papers since and before.

If you decide that the response of interest is "viewer level of interest"
in a presentation, and that there are specific things you can change
about the presentations, in such a way that you can control if a
feature/characteristic is on or off, or how much of it is present, then
you can perform a conjoint analysis.

the math is almost exactly equal to that of a classical designed
experiment.
the measurement of 'level of interest' requires some assumptions, of
which Tukey & Green explained very completely in the 1960's.  Some ref's
on request.

I've done it a number of times, with fascinating success each time.

there are always an infinite number of possible attribute changes that
could be included in such an experiment.  Your prior knowledge/insight
about videos and what is interesting about them can narrow that down to a
dozen or more, that are _likely_ to do something interesting.  Your
interest in keeping the trials within rational, executable bounds, will
force you to select 4 to 10 factors (changeable attributes).

the big problem in my view is, how are you going to expose the
respondents to the different videos?  The order of exposure may cause
serious evaluation & interpretation issues, which you will need to
address also.

BTW, 'multiple regression' is a generalized form of designed experiment,
only without the careful planning and without the ease of
interpretation.  OK, DoE is a subset of multiple regression.  OK, the
mathematical analysis is the same, but the DoE is more informative, when
it is done properly.  Now shoot me.

Jay

David Emery wrote:

> I'm a student trying to see how mulitple regression might be used to
> predict levels of interest in interactive video presentations based on
> pre-picked attributes of the content in the video.  My question is
> whether anyone has seen anything done like this before, academic or
> proprietary??  Has anyone seen this done using a different method of
> analysis?
>
> Since there are potentially millions of attributes that could affect
> interest (or some large unknown number), and only 4 are really
> examined, the expected levels of significance are very small for each
> attribute.  Is this Ok as long as the Variance Influence Factors and
> other error levels are within tolerance?
>
> Thanks,
> David
> .
> .
> =================================================================
> Instructions for joining and leaving this list, remarks about the
> problem of INAPPROPRIATE MESSAGES, and archives are available at:
> .                  http://jse.stat.ncsu.edu/                    .
> =================================================================

--
Jay Warner
Principal Scientist
Warner Consulting, Inc.
4444 North Green Bay Road
Racine, WI 53404-1216
USA

Ph: (262) 634-9100
FAX: (262) 681-1133
email: [EMAIL PROTECTED]
web: http://www.a2q.com

The A2Q Method (tm) -- What do you want to improve today?




.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to