The effects of complexity on COBOL program changes

Abstract

Due to the increasing complexity and the increasing costs of developing and maintaining computer programs, a great need exists for reliable methods of predicting the number of errors to be expected in computer programs. The capability to predict number of errors early in the software development process would be beneficial in helping to determine the amount of needed resources and the stage of completeness of a program. The purpose of this research is to determine the relationship between program changes and program complexity characteristics and to study the rate of program completeness based on changes. The hypothesis is that given two programs of equal length, the number of program changes will not be the same but will be related to complexity characteristics of the programs. The program characteristics are measured in the first version of the program. Twenty-two program characteristics are used in this study and the relationship of each one to the number of program changes is examined. The characteristics are chosen from each of Zolnowski's four categories plus a length category. Both system and programmer characteristics which have an effect on program changes are investigated. Also, the relationship of the program changes to each of the program characteristics is studied according to program size to determine how the effect of the program characteristics on the number of program changes differs as program size varies. Next, the rate at which the program changes are made is studied. The completeness rate is determined for programs of all sizes collectively and then the completeness rate for large and small programs is measured to determine if programs of different sizes have different completeness rates. The characteristic "Breaks in Flow" was found to be the best indicator of program changes. The Structure/Control Flow category was superior to the other categories for predicting program changes. Conditions not measured as complexity characteristics were found to be critical for determining program changes. Finally, approximately 75% of the program changes were made in the first two-thirds of the program runs and roughly 25% of the program changes were made in the final third of the program runs.