Print

Print


*****  To join INSNA, visit http://www.insna.org  *****

You are working with a 2-mode (bipartite) network for which you have
attributes for the vertices in 1 mode. You would like to see how the *
attributes* of the vertices in that mode are related to the *distribution* of
vertices in the other mode.

In my case, I am working with a 2-mode network in which the events are the
winning Ads in an annual advertising contest in Japan and the participants
are the Creators, members of the teams that produced the winning ads. Each
Ad has several attributes (properties that map one-on-one with the Ads).
Three of particular interest are Agency (the advertising agency that
produced the ad), Medium (TV, radio, newspaper, magazine, poster, other),
and Industry (12 industry categories into which the winning Ads are divided
for judging). Exactly one Agency, Medium, or Industry attribute applies to
each Ad.

Creators are related to Ads vis Roles. Since, however, a single Creator may
receive credit for multiple Roles, and multiple Creators can receive credit
for the same Role, these relationships are not one-to-one and cannot be
treated as attributes.

Even so, it would be interesting to know how the *distribution* of Creators
is affected by the the *attributes *of the Ads.

The problem is that, while it is simple to construct a Pajek partitions
reflecting the attributes for the Ads and use them to analyze the 1-mode Ad
network projected from an original 2-mode AdCreator network, the 2-mode to
1-mode projection removes all of the creators. And since the partition has
the same dimension as the projected 1-mode Ad network it cannot be used, as
is, to analyze the original bipartite network with its different and larger
dimension.

Consider, for example, the 1986 Ad Creator network. This bipartite network
contains a total of 1401 vertices: 480 Ads and 921 Creators. Each .clu
Partition file contains attribute code for the 480 Ads.

The command  Net>Transform>2 mode to 1 mode produces a 1-mode network of 480
Ads. The network and Partition files have exactly the same dimension=480.
But the Creator vertices in the original 2-mode network have disappeared.
But if I try to use the Partition files to analyze the original 1401-vertex
2-mode network, I get an error message, "Network and partition of equal size
needed!" So, I look for a kludge, a trick that will get me past this error.
Fortunately, there is one.

*Partition>Create Constant Partition*

allows me to create a partition of 921 vertices (921 because that is the
number of Creators I need), all of which will have a single attribute, i.e.,
"Creator," indicated by whatever integer I choose for the constant. *I have
to be careful here that this is not one of the integers already used to code
attributes in the Ad attribute partitions.

*Partitions>Fuse Partitions*
*
*
allows me to add this new constant partition to an attribute partition.
921+480=1401! I now have an attribute partition of the same dimension as the
original 2-mode network. *Here I have to be careful that the order of the
partitions to be used is correct, the original attribute partition in the
first box, the new "Creator" constant partition in the second.

When I use this technique to analyze the 1401-vertex AdsCreators network for
1986, the result of

*Info>Partition*
*
*
is

Frequency distribution of cluster values:

   Cluster      Freq     Freq%   CumFreq  CumFreq% Representative
 ----------------------------------------------------------------
         0       335   23.9115       335   23.9115 AD1_86
         1        75    5.3533       410   29.2648 AD60_86
         2        70    4.9964       480   34.2612 AD11_86
        99       921   65.7388      1401  100.0000 Ito6226
 ----------------------------------------------------------------
       Sum      1401  100.0000

which, in this particular example, means that 921 Creators were involved in
production of 480 Ads, with 75 ads produced by 1=Denstu (Japan's largest
agency), 70 ads produced by 2=Hakuhodo (Japan's second largest agency) and
335 by 0=Other (some smaller agency).

More importantly I now have a tool that lets me dig into this network in
more detail, using Ad attributes to extract subnetworks of Creators for
further analysis. But that is another trick.

*If you find this helpful, or have suggestions or criticisms to offer,
please drop me a line.*
*
*
*John*

-- 
John McCreery
The Word Works, Ltd., Yokohama, JAPAN
Tel. +81-45-314-9324
[log in to unmask]
http://www.wordworks.jp/

_____________________________________________________________________
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.