Abstract
Modular design has become a widely accepted developmental strategy to create products and systems that can be easily manufactured, upgraded and maintained. In order to achieve these benefits through improvement of a system's modularity, it must be measured. An ideal measure ought to capture modularity while being independent of other architectural factors such as size, system coupling density or the number of modules. In this work, we review past research on modularity measures. Eight modularity measures are selected for a detailed analysis. We use a design of experiments approach to analyse which metrics best measure the degree of modularity independent of other irrelevant factors. To do this, we conduct a factorial analysis of 24 canonical architectures with idealised modularity, including precisely integral, modular and bus architectures. We find that most measures produce inconsistent results, especially if the system architecture contains a bus or modules with loose internal coupling. We identify the metrics that are able to capture the degree of modularity in the most consistent manner.