Exploring Hierarchies Using the DAGMap - Methods for Multilevel Analysis and Visualisation of Geographical Networks - page 108

Fig. 7.9 The DAG shown
here describes the links
between Nestlé and its
subsidiaries. Although the
DAG is sparse, the node-link
diagram obscures the
distribution of subsidiaries
over world regions and the
ways in which higher level
companies control lower level
subsidiaries
of different commercial strategies and companies and their relationship to the
territory. Note that the visual exploration only represents a first step toward a full ex-
planation of such phenomena. Typically, users try to build hypotheses by combining
standard (text-based) exploration with Google searches. Users subsequently return
to the classical node-link visualization with questions in mind, iterating a sense-
making loop ( Thomas & Cook , 2006 ) by engaging in various intellectual postures
(from wild guesses to strong and documented assertions).
The complexity of the analysis, however, is partly due to needing to quickly
perceive the level and place of a node in the hierarchy, while simultaneously
visualizing attributes such as a company's assets and locality (country/continent) 3 .
Simply mapping a company's assets and territorial membership using node size and
color in a traditional layout (image (a) in Fig. 7.1 ) proved to be inefficient because
the node areas could not be easily compared.
The traditional layout displays at least twice as much visual information as the
DAGMap (because ancestor nodes are mapped onto colored disks as well) or edge
crossings and occlusions. Notice also that there is a lot of free space with a Sugiyama
layout, as is the case with almost all node-link layouts. Consequently, displaying all
of the nodes needed to report companies' assets requires much more space than us-
ing TreeMaps and argues against a visualization based solely on a Sugiyama layout.
Our experience with end-users clearly indicates the need to visualize the
hierarchical structure directly on the DagMap. To this end, we enable the user to
visualize how ancestor nodes cover cells in the DagMap, i.e., the user can actually
define a “strip” that contains all of the elements at a given level in the tree (which
come from the expanded DAG). In this way, the user can visualize the dependence
of elements on ancestors.
3 In fact, the original data are not exactly a DAG: there may indeed be links from a subsidiary
controlling part of an ancestor company, thus introducing cycles into the network of links.
However, this only happens exceptionally and cycles are considered marginal by our expert users,
who actually discarded them.