On Aggregating Second-Level Software Estimation Cost Drivers: A Usability Cost Estimation Case Study Javier Crespo Computer Science Dept. Carlos III University Polytechnic School Av. Universidad 30 28911 – Legan´es Madrid (Spain) jcrespo@inf.uc3m.es Miguel– ´ Angel Sicilia,Elena Garc´ıa Computer Science Department University of Alcal´a Ctra. Barcelona km.33.6 28871 - Alcal´a de Henares Madrid (Spain) {msicilia, elena.garciab}@uah.es Juan–J. Cuadrado Computer Science Dept. Carlos III University Polytechnic School Av. Universidad 30 28911 – Legan´es Madrid (Spain) jjc@sel.inf.uc3m.es Abstract Software estimation cost drivers are in many cases abstract concepts that are better estimated by breaking them down in a number of second– level aspects. This raises the need for inquiry regarding the most ap- propriate aggregation processes for them, since they significantly im- pact in the final cost estimates. Re- cent research on aggregation opera- tor design has resulted in a number of families of operators that should ideally be considered for each con- crete situation. In this paper, we illustrate this philosophy through a concrete case study in which the required usability of a given sys- tem is modeled in terms of three second–level aspects, and the OWA operator is introduced as a realis- tic and configurable summarization tool. The results illustrate the in- fluence of the design of cost driver aggregation schemes in the resulting economical decisions. Keywords: Software Cost Estima- tion, Usability Evaluation, Aggrega- tion Operators, OWA operators. 1 Introduction Existing parametric software cost estimation models consider a number of factors often called cost drivers [3] as inputs, including in most cases software size estimations. These cost drivers are of a diverse nature, encom- passing development team’s abilities, esti- mated complexity, required level of reliabil- ity of the software and schedule constraints, among others. But these cost drivers are in many cases of a somewhat “abstract” nature, so that estimating their values directly for a given project becomes a difficult task, requir- ing the development of indirect measures. For example, the influence of the overall quality and comprehensiveness of the documentation required is better expressed through the re- quirement for certain types of more specific documentation artifacts, as described in [5]. In consequence, high-level cost drivers can be broken up in more concrete and easily measur- able “second–level” aspects that contribute to them to some extent (often not precisely known). This raises the necessity for the de- sign of aggregation operators from second– level aspects to first–level ones (and possibly in some cases, from third– and further levels). Aggregation operators can be considered as mathematical objects that have the function of reducing a collection of numbers (or more generally, of values) to a unique representa- tive or meaningful one. Perhaps the most commonly used aggregation operator is the weighted mean, but this does not mean that it is necessarily always the better choice. Since a considerable amount of research has been carried out in the last decade regarding the design of aggregation operators [4], a rich ar- ray of aggregation operator families with spe- cific characteristics are available for the prac- titioner to better solve the problem at hand.