On the use of weighted sums in the definition of measures

Abstract

Several widely used functionality measures, quality models, and other measures have been defined in Empirical Software Engineering by using weighted sums. The apparent simplicity in the definition and interpretation of weighted sums make them one of the preferred ways for defining new measures. However, they have a few inherent issues that may make their definition and use quite problematic. This paper provides a critical investigation about the intuition behind weighted sums and their theoretical and empirical validation pitfalls that may make them questionable when used in the definition of meaningful and practically useful measures. Using estimation models may alleviate at least some of the problems with weighted sums.