Conclusion - Methods for Multilevel Analysis and Visualisation of Geographical Networks - page 207

position (Chap. 10 ). In Chap. 11 , Comin examined relay cities between scales based
on the assumption that most components in scientific collaboration networks were
intranational and that certain cities linked these components to the structure as a
whole. Interactions between levels influenced actors' ability to address each of these
levels, and geographical aggregation was important for decision makers. Network
patterns made it possible to identify individuals and their interactions, although
further research is needed to determine routines and patterns fully.

7 Perspectives

This volume provided an initial exploration of the application of network analysis
and visualisation to geography and introduced novel perspectives and new issues
to the field. Future work should further this approach through theoretical accounts
that link the description of networks to specific processes. For example, in which
contexts and at what scales do network patterns, such as stars or clusters, emerge and
remain robust and resilient? How are clusters and levels of organisations modified by
the evolution of the system as a whole? To what extent does geographical distance
contribute to the development of different networks at different scales? Adapting
the methods presented in this volume to make them accessible to geographers will
allow them to address these issues and many others.
It is worthwhile to review the origin of the project that brought the authors in
this volume together. Many problems that were addressed in the present volume
had never been fully visualised, and the case studies presented here often involved
data sets that had never been fully explored because the data were too extensive
to be spatially mapped. The interactive visualisation of graphs provided a means
to analyse the entire data set based on the visual analysis and inspection of salient
features.
One consequence of this approach was the realisation that physical distances
were not central to the analysis, which was based on topological properties of
the data sets. This pragmatic approach contrasted physical distances with “social”
distances that were measured through the intensity of interactions between entities.
This idea is a fundamental principle of network science that parallels similar con-
cepts in the social sciences ( Martin , 2009 ), and this perspective allowed computer
scientists to contribute to the data analysis. The present volume documents the value
of this type of idea exchange.
The graph visualisation methodology also exploited the central paradigm of
“scale”, which generated many debates on network hierarchies and the “small
world” phenomenon. The constraints imposed by the manipulation of large datasets
on standard visualisation displays drove downsizing approaches that transformed
large graph structures into hierarchies of smaller subgraphs. Because hierarchies
occur as a structural principle in nature ( Pumain , 2006 ), many computational
approaches have been successfully applied to these datasets to render the entire data
set accessible; in addition, the downsizing process implicitly addressed the concept