A Systematic Approach to the Analysis of Function Point - COSMIC Convertibility

Objective. After the COSMIC FSM method has been proposed, the issue of convertibility between traditional function points and COSMIC function points has arisen. Several studies have been performed, in order to evaluate whether a correlation between traditional and COSMIC FP measures can be established. Most of these studies applied linear regression analysis to sets of data concerning the size of projects measured in both FP and CFP. These datasets are often characterized by skewness, outliers and heteroscedasticity: it is therefore necessary to evaluate the statistical validity of the studies that used such datasets. Method. Statistical analysis techniques for dealing with the aforementioned problems have been proposed. Here we employ log-log transformations to normalize the datasets, and least median of squares regressions to decrease the dependence on outliers. The resulting models are compared using rigorous statistical methods. Results. The paper shows that different datasets tend to produce different models of the relationship between COSMIC and traditional function points. Moreover, different datasets are best modeled by different types of regressions. Conclusions. On the base of the collected evidence it seems impossible not only to claim that there is a unique formula that can be used to convert traditional function point measures into COSMIC function point (or vice versa), but even that a given type of regression yields the best models for every possible dataset. In many cases, however, robust regression techniques yield reasonably precise conversion formulas.