SOCNET@LISTS.UFL.EDU

View:

 Message: [ First | Previous | Next | Last ] By Topic: [ First | Previous | Next | Last ] By Author: [ First | Previous | Next | Last ] Font: Monospaced Font

Subject:

Re: MDS visualization of CONCOR output

From:

Date:

Wed, 31 Mar 2010 18:09:37 +0200

Content-Type:

text/plain

Parts/Attachments:

 text/plain (84 lines)
 ***** To join INSNA, visit http://www.insna.org ***** Hi, Even though the Concor algorithm can provide results that are intuitive, there has been quite a lot of critique raised against it (e.g. Sim and Schwartz 1979; Doreian 1988; Faust 1988). One of its founders has explicitly said that there is "[no] justification for advocating the iteration of [correlance coefficients] matrices as a method for analysis of data" (Schwartz 1977:266ff). Tha algorithm always produced 2 splits, and makes subsequent partitions quite arbitrary when it comes to which of these splits to split further. If there are strong arguments to use structural equivalence, perhaps it would be better to simply use the correlance matrix (distance-based or Pearson) and use MDS to plot these in two dimensions, subsequently interpreting this in combination with a dendrogram. Depending on the data, and what you are looking for, you should also look into regular equivalence (if that specific sociological notion of "roles" is applicable to your particular study). I might be wrong, but it was quite some time since I last saw a solid argument on the advantages of using the Concor algorithm. Yours, Carl Semenov Alexander wrote: > ***** To join INSNA, visit http://www.insna.org ***** > > I guess you have your data already clustered by CONCOR, and you > would like > to present the reduced blockmatrix by MDS? > > When I run MDS directly in NetMiner it draws me a map with only a few > visible actors. It seems like 90% of nodes are exactly in the centre > and one-two are at the end of each axis. It's almost imposible to > interpret such picture. But when I check MDS as an output option in > CONCOR, it shows me nice scatter plot with 3 distinctive axis and some > dispersion. This picture corresponds with my substantive assumptions > and very interesting for interpretation. > > Anyhow, you will be much more exposed to criticism for adopting > CONCOR, than > for using MDS... > > Why? Unfortunatelly, I really doubt it =( Here in Russia there is only > one thesis on SNA and only one book about it. And the author of both > of them is my tutor. My search for SNA in local sociological journals > revealed only 4 articles. So, this listserve is the only place for me > to search an advice. That's why I'm asking such a noob questions. > Alexander. > > > 2010/3/31 Balazs Vedres <[log in to unmask] > > > ***** To join INSNA, visit http://www.insna.org ***** > > "rarely used" - I think it is very often used. > I guess you have your data already clustered by CONCOR, and you > would like > to present the reduced blockmatrix by MDS? That is fine. > You can also use MDS to represent the original structural > similarity measure > matrix, and possibly include the centroids of CONCOR blocks. > Anyhow, you will be much more exposed to criticism for adopting > CONCOR, than > for using MDS... > > -----Original Message----- > From: Social Networks Discussion Forum [mailto: > -- Carl Nordlund, BA, PhD student carl.nordlund(at)hek.lu.se Human Ecology Division, Lund university www.hek.lu.se _____________________________________________________________________ SOCNET is a service of INSNA, the professional association for social network researchers (http://www.insna.org). To unsubscribe, send an email message to [log in to unmask] containing the line UNSUBSCRIBE SOCNET in the body of the message.