A couple of "newer" ways of analyzing screening designs (not just P-B
variety) include Box & Meyer's "Bayes plot" (1996 Technometics, I believe)
and Lenth's PSE (also Technometics, forgot the year). Lenth's PSE can be
computed fairly easily.
I believe Bert's approach is similar in spirit to one iteration in Wu &
Hamada's stepwise procedure. However, as Bert said, to get more mileage out
of the main effect plans, you need to be willing to make more assumptions
that may or may not be appropriate.
Cheers,
Andy
> -----Original Message-----
> From: Gunter, Bert [mailto:bert_gunter@merck.com]
> Sent: Wednesday, October 22, 2003 10:51 AM
> To: 'Spencer Graves'; Firat ÖZDEMIR
> Cc: s-news@wubios.wustl.edu
> Subject: Re: [S] plackett burman design
>
>
> All:
>
> (Non-geometric) Plackett-Burman designs have the very
> interesting property that they are "3-projectable." This
> means that for ANY subset of 3 variables, the design
> contains a full factorial (+ one or more 1/2
> fractions) in those 3 variables.. This was proven during the
> 1990's by Wu, Lin, Box, Tyssedal, Chen and colleagues in a
> series of papers. This means that there is an alternative
> method (and philosophy of) analysis: Consider fits in all
> possible subsets of 3 variables and pick the (few?) best.
> This allows for higher order interactions under the
> assumption of parsimony -- that only up to 3 of the variables
> are needed to give a good fit. Box and Tyssedal discuss this
> in their paper (about 1996 , I think). I have in fact used
> this approach for screening designs and it generally works
> quite well (though not infallibly, of course, as (partial)
> confounding rears its head in other ways).
>
> Cheers,
> Bert Gunter
> Biometrics Research RY 33-300
> Merck & Company
> P.O. Box 2000
> Rahway, NJ 07065-0900
> Phone: (732) 594-7765
> mailto: bert_gunter@merck.com
>
> "The business of the statistician is to catalyze the
> scientific learning
> process." -- George E.P. Box
>
>
> -----Original Message-----
> From: Spencer Graves [mailto:spencer.graves@pdf.com]
> Sent: Wednesday, October 22, 2003 10:17 AM
> To: Firat ÖZDEMIR
> Cc: s-news@wubios.wustl.edu
> Subject: Re: [S] plackett burman design
>
>
> Have you tried searching "www.google.com" for "Plackett-Burman
> design"? I just got several hits there.
>
> Most modern books on experimental design, e.g., Box, Hunter,
> Hunter (1978) Statistics for Experimenters (Wiley) describe
> Plackett-Burman designs and cite Plackett and Burman's
> original paper.
> As far as I know, the standard analysis involves computing
> coefficients
> for a saturated model (or "effects" = twice the corresponding
> regression
> coefficient) and making a normal probability plot of those numbers.
> Points obviously off the line are considered to be statistically
> significant. A gap in the middle or an offset from mean 0
> indicate an
> outlier (Box on Quality and Discovery, Wiley, 2000). Wu and Hamada
> (2000) Experiments (Wiley) recommend an advanced stepwise
> regression.
> I've developed S-Plus code for my version of advanced stepwise
> regression, downloadable from "www.prodsyse.com".
>
> Hope this helps. spencer graves
>
> Fırat ÖZDEMİR wrote:
>
> > is there any paper, web page or something else available where I
> > can have detailed information on analysis of
> plackett-burman designs
> >
> > firat.ozdemir@deu.edu.tr <mailto:firat.ozdemir@deu.edu.tr>
> >
>
>
> --------------------------------------------------------------------
> This message was distributed by
> s-news@lists.biostat.wustl.edu. To unsubscribe send e-mail
> to s-news-request@lists.biostat.wustl.edu with the BODY of
> the message: unsubscribe s-news
>
>
> --------------------------------------------------------------------
> This message was distributed by
> s-news@lists.biostat.wustl.edu. To unsubscribe send e-mail
> to s-news-request@lists.biostat.wustl.edu with the BODY of
> the message: unsubscribe s-news
>
|