An exploratory study on a linear model for measuring software quality

A pilot study was conducted to develop a linear regression model to measure software quality. The model incorporated four quality components representing efficiency, understandability, modifiability, and implementation of requirements. Halstead measures were used as regressors along with a programmer rating. The study used a group of second-semester Fortran programming students who implemented the same program individually. A series of standard tests was utilized to measure the quality components. The results showed linear dependency of all quality components, except understandability, on Halstead metrics. The efficiency, modifiability, and requirements components were expressed using only volume, level, and programming time metrics. Thus, those three measures appear as the most promising for validation research. However, the best correlations for each quality component were given by the programmer rating, suggesting that nonlinear solutions might be more appropriate