Stensrud and Myrtveit present a critique of previous work on the evaluation of the productivity of software development projects, and advocate a technique called data envelope analysis (DEA). Their principal criticism of previous work is that it is based on the assumption that there is a linear relationship between product and effort (constant returns to scale). The authors argue that most sets of projects will exhibit variable returns to scale. They also maintain that previous studies are flawed because they ignore significant drivers, such as infrastructure development, the business environment, and the fact that complex projects may generate several types of outputs.
The DEA technique is based on the assumption that the efficient projects in a data set are those located on the convex hull of the data points in (N+1)-dimensional space (the “frontier”), where N is the number of types of products. A peer index is computed and used to identify a set of efficient projects, to which certain other projects can meaningfully be compared. The method is illustrated using two data sets: the classical data of Albrecht and Gaffney [1] and a set of 30 projects involving the SAP enterprise resource planning (ERP) software product. The latter provides the title for the paper, however the method is general.
The authors’ explanation of DEA is uneven: superficial in some areas, with excessive detail in others. Their criticism of previous work is argumentative and repetitious.