Using Group and Subsystem Level Analysis to Validate Software Metrics on Commercial Software Systems

This paper reports the results of a study which examined the relationship between a collection of software metrics and the development data (such as errors and coding time) of three commercially produced software systems. The software metrics include both measures of system interconnectivity and measures of system code. This study revealed strong relationships between the metrics and the development data when individual components were aggregated by structure (into subsystems) or by similarity (into groups). The subsystem and group results imply that research and application of metrics should be focused above the component level. The group results also imply that metrics can guide the effective application of project resources by identifying those groups which, for example, will contain a disproportionately large fraction of errors. Finally, the study showed the overall utility of two interconnectivity metrics: Henry and Kafura's information flow metric and McClure's invocation metric. This result is significant because interconnectivity metrics can be applied early in the life cycle.