Lohr, Claire L. (1997, American Society for Quality) Lohr Systems, Reston, VA
Direct and indirect quality indicators were used to compare the characteristics of two large software systems at IBM Software Solutions Toronto Laboratory. The two relational database management systems were a legacy system (LS) in place since about 1976 and a new system (NS). The systems had 1302 and 995 modules, respectively. The direct quality indicator was a metric for defect fixes. The indirect indicators consisted of 16 metrics in four areas: design (spanning six metrics, including module level system complexity); size (two metrics on lines of code and comment lines); change (two metrics on changed source instructions and software maturity index); and complexity (six metrics, including referenced declarations.) Results indicated that LS had 2367 defect fixes and NS had 7824, the large apparent difference being accounted for by analyzing modules as new, mature, or evolving. The defect fixes were unevenly spread over both LS and NS modules. Indirect metrics produced comparable results, except in the high level design complexity metrics. This difference could be explained by considering the object-oriented and structured methods used in NS. A correlation analysis of the direct (defect) metric and the indirect (code) metrics resulted in no strong correlation findings, the highest correlation being the 0.731 rank correlation between defect fixes and changed source instructions. Linear regression analysis and modeling with logarithmic data transformations produced highly skewed results that were inadequate for predicting defect fixes from the indirect metrics.
Case study,Correlation,Defects,Software quality,Metrics,Regression analysis,Measurement and control