Date: Mon, 7 Jul 2008 19:46:49 -0500
Reply-To: paul@WUBIOS.WUSTL.EDU
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Paul Thompson <paul@WUBIOS.WUSTL.EDU>
Subject: Re: PROC MDS using preference rankings
In-Reply-To: <a2e4b317-9093-483c-9cb4-f51f20d2fcc5@25g2000hsx.googlegroups.com>
Content-Type: text/plain; charset=ISO-8859-1; DelSp="Yes"; format="flowed"
Quoting jenmoocat <sollje2002@YAHOO.COM>:
You can go 2 ways with this.
1) construct a similarity matrix from your preference rankings
2) do an unfolding type analysis for your preference rankings
Regardless, I would include all points in the main analysis, and
compare the results in some manner after scaling.
> Ever have one of those days where you feel like you are just missing
> something?
> Hopefully someone can help -- because I am beginning to feel like an
> idiot.
>
> I have preference rankings for 6 brands for two time periods (t=0 and
> t=1).
> I would like to overlay the preference map from time t=0 with the map
> from time t=1, so we can easily see how preferences have changed over
> time.
>
> My data looks like the following (for one time period and 5000
> respondents):
>
> respondent brand1 brand2 brand3 brand4
> brand5 brand 6
> 1 5 2 1
> 2 3 5
> 2 4 3 3
> 3 3 3
> 3 2 2 5
> 1 1 1
> 4 1 2 4
> 4 4 4
> ...
> ...
> ...
> 5000 2 5 5
> 3 2 1
>
>
> And I have another dataset for the next time period --- with changed
> rankings (hopefully).
>
> I've spent the past couple of days reading the PROC MDS documentation
> pages and the MDS chapter in "Multivariate Data Analysis" by Hair, et
> al., and searching around the web --- and I am stymied.
>
> Although the textbook talks about being about to use preference
> rankings, all examples that I've seen (in the book and on the web)
> have been based on starting with a similarity matrix. And I just
> don't quite grok the best way to go from my preference ranking data to
> a similarity matrix. Or whether PROC MDS can be used with raw ranking
> data, instead of the matrices.
>
> Should the similarity matrix just simply be constructed using the
> following steps:
>
> for each respondent, calculate a matrix that contains the simple
> numerical difference between the rankings:
> so for respondent 1, it would look something like:
>
> brand1 brand2 brand3 brand4
> brand5 brand6
> brand1 0 3 4
> 3 2 0
> brand2 -3 0 1
> 0 -1 -3
> brand3 -4 -1 0
> -1 -2 -4
> brand4 -3 0 1
> 0 -1 -3
> brand5 -2 1 2
> 1 0 -2
> brand6 0 3 4
> 3 2 0
>
>
> then either 1) create one "similarity matrix" by taking the averages
> over all responders or
> 2) stack these matrices, one responder over another (and use the
> CONDITION=ROWS option)
>
> Am I on the right track? I feel like I am just missing something
> obvious and am at the point where I feel like banging my head against
> the wall.
>
> Can anyone shed a little light on this for me?
>
> Thanks muchly in advance,
>
> -jennifer
>
| 
