An experiment to assess task complexity in the supervision of networked systems

Supervisory monitoring of large-scale networked systems is an area of continuing research interest. Previous studies suggest that a person´s working efficiency is influenced by the `complexity´ of the networked system. However, a useful quantitative measure of such complexity has been elusive. The present study postulates that cyclomatic graph complexity may offer a practical step forward in this direction. This metric has been applied in software engineering with a degree of success in certain circumstances. An experiment was conducted to test this hypothesis. Participants were presented with a set of abstract graphical displays of simulated network flow information. Numeric data items were shown for each node, their values being routinely updated to mimic dynamic flow behavior. Several different cyclomatic complexity levels were used. The task was to locate the nodes where specific patterns of abnormal network activity seemed to be occurring. Decision time and task accuracy were the major dependent variables used to assess human performance. Subjective ratings of individual task difficulty were also evaluated. Statistical analysis of the results shows that the complexity metric may have some useful applicability in predicting these measurements