Issue No. 06 - November/December (1998 vol. 10)
<p><b>Abstract</b>—This paper relates four objective measures of program structure and three subjective ratings of program complexity to the number of postrelease documented errors contained within 80 commercially developed Prolog programs. All seven measures show a significant correlation with the number of errors. A factor analysis showed that the objective and subjective measures were indeed different measures, although a hierarchical analysis of oblique factors showed a strong common root. Finally, the Mann-Whitney <it>U</it> test was used to determine whether the measures could differentiate between those programs with errors and those with no documented errors. The results suggest that "complexity" can be grounded in terms of the difficulty to debug or test a program, while measures of "structure" require a detailed count of the number of predicates used within the program.</p>
Complexity measures, program structure, error-proneness, system quality, rule-based systems, empirical, Prolog.
T. T. Moores, "Objective vs. Subjective Measures of Error-Proneness for Rule-Based Programs," in IEEE Transactions on Knowledge & Data Engineering, vol. 10, no. , pp. 1008-1012, 1998.