Date:         Thu, 22 Sep 2005 09:37:12 -0400
Reply-To:     emaguin@acsu.buffalo.edu
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         emaguin@acsu.buffalo.edu
Subject:      Re: 2nd Attempt: Grouping Zip Codes
In-Reply-To:  <E08B564BAA29FE4F8EC8BAF42FE0A01F126717@bubsex02.cen.csin.cz>
Content-Type: text/plain; charset=us-ascii

As i've been watching this conversation develop, i've come to think there
is a practical solution to this. The data is a square matrix, 0's on the
diagonal and symmetric about the main diagonal. Pick a distance. Go row by
row through the matrix and 'blank' out elements in the row that are over
the distance threshold. Now go column by column through the matrix and look
down the column. I think i'm left with three conditions. A column will have
a) exactly one non blanked element, b) multiple nonblanked elements, or c)
all elements blanked. Condition a) is perfect because each zip code is
assigned to just one 'center'. Conditon b) is one problem. But, practically
does it matter. If, for instance, the underlying motivation for the
question were establishing distribution or service centers, why not
randomly assign the area to one of the eligible centers? Condition c) is
the other problem because the zip code is 'out of reach' of all centers. So
i either increase the distance and rerun the job or simply assign the zip
code to the closest center.

Can all this be done in spss? I think so. But, it can't be done in a closed
form, single pass solution.

Gene Maguin