If we are using the node-link visual metaphor to represent relational data we soon reach the limitations of this kind of graph visualization. If the graph becomes dense, i.e., has many links the probability of link crossings increases. These crossings cause visual clutter, a state in which an excess of items or their disorganization leads to a degradation of performance at some task as stated by Ruth Rosenholtz.
Actually, layout algorithms have been developed and implemented with one goal to position the vertices with respect to reduce link crossings. Also an edge bundling approach may be used to reduce visual clutter but single graph edges are hard to trace by the eye.
In our work, we ask the question if node-link diagrams are still readable, understandable, and useable when the links are not drawn in its full length. To find this out, we conducted a user study with 42 participants comparing traditional and tapered link styles, three different graph sizes, and six different link lengths. We ask three different tasks and measured accuracies and completion times.
The outstanding result of this study is the fact that if the link length is roughly at 75 percent of its original length participants are much faster when answering the task and they are also more accurate in many cases. If you are interested in our work, please read our paper presented at the 19th International Symposium on Graph Drawing in Eindhoven, The Netherlands or write an email. Once we found out, that partially drawn links are also useful to some degree we work for a graph visualization tool that uses this as an interactive visual feature.