Performance of layout algorithms: comprehension, not computation

Many algorithms address the problem of rendering an abstract graph structure as a diagram in as efficient and as elegant a manner as possible. The criteria for judging the worth of these algorithms are typically the extent to which they conform to common aesthetic criteria (e.g. minimising the number of crossings, maximising symmetry), or their computational efficiency. The algorithms are not usually judged on their ability to produce diagrams that maximise humans' performance on tasks which require their use. This paper presents an example experimental methodology for considering the relative worth of eight layout algorithms with respect to human performance, together with details of an experiment using a single graph. The results indicate that, with the exception of one algorithm, there is no statistical difference between the performance data of the algorithms when applied to this graph, indicating that they produce drawings of comparable difficulty. This result is despite the different aesthetic bases for the algorithms.