FAULT SEVERITY IN MODELS OF FAULT-CORRECTION ACTIVITY

This study applies canonical correlation analysis to investigate the r elationships between source-code (SC) complexity and fault-correction (FC) activity. product & process measures collected during the develop ment of 91 commercial real-time product provide the data for this anal ysis, Sets of variables: represent SC complexity and FC activity, A ca nonical model represents the relationships between these sets. s-signi ficant canonical correlations along 2 dimensions support the hypothesi s that SC complexity exerted a causal influence on FC activity during the system-test phase of the real-time product, Interpretation of the s-significant canonical correlations suggests that two subsets of prod uct measures had different relationships with process activity. One is related to design-change activity that resulted in faults, and the ot her is related directly to faults. Further, faults having less impact on the system-test process associated with design-change activity that occurred during the system-test phase, while those having more impact associated with SC complexity at entry to the system-test phase. The study demonstrates canonical correlation analysis as a useful explorat ory tool for understanding influences that affected past development e fforts. However, generalization of the canonical relationships to all software development efforts is untenable since the model does not rep resent many important influences on the modeled Latent variables, eg, schedule pressure, testing effort, product domain, and level of engine ering expertise. Work remains to specify subsets of indicators and dev elopment efforts for which the technique becomes useful as a predictiv e teal.