Garry, your work sounds very interesting and I look forward to seeing your
One set of techniques for handling geographical effects is spatial analysis
and spatial regression. I took a workshop on spatial analysis with Luc
Anselin about a year ago. Spatial analysis uses, among other techniques,
distance matrices and spatial autocorrelation where values at one node are
influenced by values at "nearby" nodes.
There clearly is room in these techniques for using social distances derived
from network data in place of or in addition to geographic distances. I
haven't had a chance to try out any of the spatial regression techniques and
I don't remember seeing any network analyses using these techniques. Has
anybody tried these yet?
----- Original Message -----
From: "Garry Robins" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Thursday, September 05, 2002 10:17 PM
Subject: Fwd: Probably Naive But ...
> Responses to the original query have pointed to some really good empirical
> work, but there is an interesting challenge here for the modeling of
> networks. Seems to me that we do not have a very good handle on models
> incorporate both network data and spatial information concurrently. One
> the problems may be that the relevant spatial information should not
> necessarily enter models in a typically Euclidean way - to pick up an
> example already raised, maybe what counts is that I commute through five
> subway stations, irrespective of whether the actual distance is five or
> kilometres. Pip Pattison and I have a piece in the 2002 Sociological
> Methodology that discusses what we call social settings. These are sets of
> nodes and/or possible network ties that can be used as a modeling device
> various ways. Settings can be construed quite broadly as social groupings
> or possibly as regions of geographic space, and they can also overlap one
> another. At this year's Sunbelt, Pip illustrated how inclusion of spatial
> settings information in the form of Florentine neighbourhoods added to the
> modeling of John Padgett's Medici data. We hope to write this up, together
> with some other empirical settings examples, soon. This is our current
> shot at modeling space and networks simultaneously. (We're planning to
> this general approach into a large scale empirical study of rural personal
> networks, which also bears on the original query - but we haven't
> much data as yet.) I'm well aware that others have also been thinking
> network/spatial modeling issues, and it would be useful to hear of current
> work of relevance.
> It raises another point of interest - to what extent does a single network
> in itself provide sufficient information for us to draw clear conclusions?
> To what extent do we need to collect additional information (spatial,
> settings, multiple time points, etc) to have confidence about our
> inferences regarding the network? Of course, much depends on the research
> question and, naturally, sometimes we just make do with what we have. Even
> so, seems to me that we often do not have a very clear idea about the
> extent of the information, or other types of observations, that we should
> be seeking in network studies.
> Garry Robins