A Large-Scale Empirical Comparison of Object-Oriented Cohesion Metrics

Cohesion is an attribute of software design quality for which many metrics have been proposed. The different proposals have been made largely on theoretical grounds, with little evidence of actual use. This makes it difficult to provide advice to software developers as to how to interpret the measurements any given metric produces. This paper presents the first large-scale empirical study of object- oriented cohesion metrics. We apply 16 metrics from the literature, as well as a number of variations, to 92 open source and industry Java applications ranging in size from a few classes to several thousand, over 100,000 classes in all. Our results show that by and large applications have similar distributions of measurements according to any given metric, but that the distributions can be quite different across metrics. This provides useful information for the ongoing empirical validation efforts for cohesion metrics.