A Model for Defining Coupling Metrics

Many metrics have been proposed to measure coupling—the degree of association between modules in a system. However, most metrics are under-defined, meaning that different tool developers can reasonably implement the same metric in many ways. This gives rise to families of metrics, which are superficially similar but potentially produce different results. To understand how different these metrics are, we propose a single model of coupling based on the concept of dependencies. This model is useful for defining existing coupling metrics, analysing their differences and clarifying divergent implementations. We demonstrate its efficacy by using it to describe existing coupling metrics and inform tool development. We have applied the tool to the 112 systems in the Qualitas Corpus, generating 21 million measurements from 88 coupling metrics. The simplicity of the tool implementation and the number of metrics it supports demonstrates the usefulness of our model.