Re: Design of Experiment

[Tim Folkerts <tfolkert@FHSU.EDU>]

I wasn't trying to imply that factorial designs are the only approach or
are appropriate for every situation.  I was merely trying to inform that
there exist formal approaches that can be much more effective than
intuition.  Of course, there is always a balance between previous
understanding, what you are trying to accomplish, and the method of
experimentation you apply.

I chose an example with factorial designs because they are relatively
straightforward to understand.  DoE has many approaches, designed for
different situations - screening for main effects, screening for
interactions, maximizing a parameter, optimizing a parameter, etc.

> Some people consider this approach (#3) to be "the" big idea
> that you get from DoE.  I'm not sure I agree;  I attack with both
> edges of a two-edged sword:
> -- there are other good ideas in DoE, and
> -- the "factorial analysis" (partial or otherwise) idea isn't
>     nearly as useful as some partisans seem to suggest.

Absolutely.

> To the extent that DoE suggests you can attack hard problems without
> any understanding, it is a Bad Idea.

Absolutely.

But there are two levels of understanding understanding something of the
problem, and understanding something about attacking the problem.
Physicists tend to be quite sophisticated about trying to understand the
problem, but then simplistic when choosing what parameters to vary by how
much and in what order.


> Someday you might be lucky enough to be presented
> with a problem that is "just barely intractable" so that
> a trick like partial factorial analysis makes a real difference,
> but I wouldn't plan on it if I were you

Unless each experiment involve shutting down an assembly line and the boss
doesn't really want to do that for 64 hr at $10k/hr if 16 hr will give him
the same information.  Or each experiment takes a day to complete, and
there are only 20 days until the end of the semester.  I think the "real
world" is full of such situations.


> My view of the principles goes like this:
> ++ The future lies in between:  what we need are principles
> and systematic methods for capturing _partial_ understanding,
> and then doing a search guided by this partial understanding.
> This is a topic that is ripe for good research -- research
> into the principles, as opposed to just adding to the
> bestiary of ad-hoc heuristics.

At it's best, this is exactly what DoE is trying to do.



Tim Folkerts


 Fort Hays State University
 Hays, KS  67601
 785-628-4501

This posting is the position of the writer, not that of SUNY-BSC, NAU or the AAPT.