A study of macro level complexity metrics

Graduation date: 1986Controlling the "complexity" or "understandability"\ud of computer software is important because of its impact on\ud program testing and maintenance. Of the large number of\ud complexity metrics that have been developed to measure the\ud complexity of a computer program, most assess the\ud "micro-complexity" of each subprogram and few assess the\ud "macro-complexity" of the entire program. The dissertation\ud introduces a new macro-complexity metric that incorporates\ud global variable and parameter usage, average module\ud micro-complexity, and internal documentation.\ud Validation, or demonstrating that a metric does work,\ud is a difficult problem. Industry is seldom willing to\ud provide researchers with actual source code because of the\ud fear of it falling into the hands of competitors. The\ud dissertation presents a method, called a Reduced Form,\ud that allows researchers access to the information about\ud the source code, but prevents reconstruction of the code from the information.\ud In a field study that compared the new\ud macro-complexity metric with several other macro and\ud popular micro metrics, the Reduced Form was used to\ud collect empirical data. The results of the study suggest\ud that the new metric performs significantly better than any\ud of the others studied. However, a simple count of lines of\ud code, while not as highly correlated with program errors\ud as the new metric, does prove to be much easier to\ud compute, and performs at least as well as any of the other\ud metrics studied with the exception of the new metric.\ud Therefore lines of code probably remains the metric of\ud choice for most situations