Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Frequency weighted cluster analysis

From   Nick Cox <>
Subject   Re: st: Frequency weighted cluster analysis
Date   Wed, 11 Jan 2012 08:20:47 +0000

Good point. Reminds me of what is so unsettling about cluster
analysis, that quite contradictory assumptions are often on offer.
That does not arise elsewhere in statistics!

(Well, yes it does!)


On Wed, Jan 11, 2012 at 12:49 AM, Brendan Halpin <> wrote:
> On Wed, Jan 11 2012, Nick Cox wrote:
>> Why would a cluster analysis change because some observations are
>> duplicated? The similarity or dissimilarity of objects is not affected
>> by their frequency. What does this SAS statement do that should be
>> replicated by Stata?
> It depends on the linkage used. For linkages where the size of the
> clusters matters in choosing which pair of clusters to agglomerate next,
> having one versus many of a particular case will change the results.
> Thus it doesn't matter for single or complete linkage, but it does for
> Ward's.
> I'm speculating that it's possible to write the low level code to take
> advantage of the duplication. I'm guessing that the SAS command (which I
> have only seen online -- I don't have access to SAS) exploits this
> possibility.
> At the moment I'm just fishing for clues -- I have an idle idea[1] for
> an analysis that would exploit high levels of duplication.
> Regards,
> Brendan
> [1] Procrastination, in other words.
*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index