Networks and link data, was preprint version "Globalisation in science in 2005"
David E. Wojick
dwojick at HUGHES.NET
Sat Mar 10 11:27:16 EST 2007
Thank you Christina, very useful.
Most network analysis is quite rightly focused on networks for which
we have link data. I am interested in, and took Loet to be talking
about, some more basic network of scientists, about which we have
very limited data. It is this concept, that there is an existing
network to be discovered, that drives us to seek new kinds of data.
More especially, I am trying to study diffusion of thinking among
scientists at a level for which there may be no individual link data.
Message by message transactions if you like. This is analogous to
diffusion of groundwater, where we know we cannot track individual
molecules. In the case of water we monitor pressure using a 3D grid
of sensors. In my case I am looking at the migration of new language
in, say, journals, or proposals, or conference proceedings.
The math of discrete network analysis may or may not apply in my
cases. How to bridge the gap between discrete network analysis and
diffusion seems to me to be an important question. I think a similar
problem arises in statistical mechanics.
For my own learning process, let me try to respond to your original
questions, and then maybe the more experienced on the list can
>My first question is, does every initial transaction between two
>people, no matter how slight, establish a link in the network? The
>alternative might be that certain thresholds of interaction must
>apply, such as co-authorship or citation. My second question is, do
>you think that the strength of a link might usefully be measured by
>the amount of information that passes between the people linked?
Network analysis works on relationships as the basic unit (not
attributes, for example). How these relationships are
defined/determined depends on what makes sense for the system being
studied. We can look at co-authorship, co-citation, co-membership
(in an organization) or a 2-mode thing like co-attendees of an
event... So we might look at co-authorship of only peer reviewed
journal articles that are in journals indexed by ISI -- which sets
the bar fairly high. For things like co-authorship, the relationship
is non-directional but can be valued. It can have different weights
for the number of times the authors appear together if that makes
sense to the phenomenon you are studying. Many times, it makes more
sense to just have 1s and 0s -- link or no link. As for amount of
information, one of the problems in network analysis is obtaining the
data. Can you interview everyone involved? How much information can
you obtain about the actors? Can that information be easily made
into a matrix for mathematical operations? I think there's the
pragmatic aspect of what is do-able as well as the science aspect of
what you want to look at.
The standard book on SNA is Faust and Wasserman and there have been
several relevant chapters in ARIST on citation analysis, but you
might do best by reading Dr. Leydesdorff's books and web page.
Christina K. Pikas, MLS
R.E. Gibson Library & Information Center
The Johns Hopkins University Applied Physics Laboratory
Voice 240.228.4812 (Washington), 443.778.4812 (Baltimore)
"David E. Wojick, Ph.D." <WojickD at osti.gov>
Senior Consultant -- The DOE Science Accelerator
A strategic initiative of the Office of Scientific and Technical
Information, US Department of Energy
391 Flickertail Lane, Star Tannery, VA 22654 USA
http://www.bydesign.com/powervision/resume.html provides my bio and
presents some of my own research on information structure and
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the SIGMETRICS