Talk:Cohesive blocking
From InterSciWiki
Hi Folks - I ran the SAS version on this net: Took about 22 secs to run; highest K is 3. I'll post the picture of the blocking tree and the sociogram once I figure out how to get it up here in decent resolution...
Best, Jim
[edit] Cross-talk
Wouter de Nooy, Andrej Mrvar and Vladimir Batagelj, Exploratory Social Network Analysis with Pajek, CUP, 2005. Amazon. ESNA page.
[edit] More from Jim Moody
There are two versions of nestedness (which is keyed in the footnotes of the orig. paper). Either the level in the cut structure (main - as you describe) or the highest k-value of the components they belong to. The latter has the advantage of being algorithm independent and easily interpreted (it can be translated directly to numbers of cycles, independent paths, etc.). Getting the k-level is simplest, since that's already tagged to the output vector.
The depth-in-tree version is more fine-grained than the highest k-level, and that is why i like it. It is also a graph-relative measure, so it's specific to a particular graph (so being nested 4 levels in two different networks might mean very differnet things). So that's something to keep in mind.
To get the depth in the tree, you have to use a depth algorithm. There is a SAS IML routine called "depth.mod" in my span set, or you can use the partion>depth in PAJEK. Apply these to the tree graph that results from the cohesive blocking, and you then get a depth level for each cut. This then becomes a simple indicator for each partition. So it would conceptually be something like:
p1 p2 p3 p4 p5 p6 [zero-one mat indicating row node is in part px]
and then a value for the depth. Then you just seach for each node which partition they belong to, and then from there pick the maximum of the corresponding values.
The code folows the logic of identifying "deepest" in the sample program.
I've attached a SAS program that does this. It's the Gangon example, with a new part at the bottom. This is the code you want.
Pts,jim
