A study of software metrics

This article reports on an empirical study conducted to explore the interdependencies among a number of software metrics and validate a new metric. Also a model for code faults in terms of static metrics is presented and investigated. The software analyzed is a commercial data-base system currently being sold and used widely. Availability of various post-release versions of this production-quality software package, together with the corresponding history data, provide a new dimension for metrics analysis. The results of this study indicate that: 1. 1) all of the static metrics considered correlate well to each other and also to the size metric; 2. 2) the residual complexity metric was validated against other static metrics; 3. 3) none of the metrics under consideration was able to identify all of the variations in code faults; in fact, even a multiple-variable model accounted for only one quarter of the variance in the number of code faults; and 4. 4) the static metrics' intercorrelations are consistent across versions.