Characterizing a data model for software measurement

In order to develop or acquire a software product with appropriate quality, it is widely accepted that quality must be identified, planned, measured and controlled during the development process using quality measures based on a quality model. However, few practitioners in the software industry would call measurement a success story. This weakness arises, on one hand because the people involved are not always aware of the importance of collecting measures. The policy of the management board must make sure that people are sufficiently motivated and that data is actually collected in the specified way. On the other hand, software measures have been often poorly defined in industry. When software measurement definitions are incomplete and/or poorly documented, it is easy to collect invalid or incomparable measures from different data collectors. Thus, the primary issue is not only whether a definition for a measure is theoretically correct, but that everyone understands what the measured values represent. Then, the values can be collected consistently and other people, different from the collectors, can interpret the results correctly and apply them to reach valid conclusions. The objective of this paper is to present a data MOdel for Software MEasurement (MOSME) to explicitly define software measures, providing the elements required to describe a consistent measurement process. MOSME can be used for defining and modeling data sets of software products involving several software projects. The inspiration of this work comes from the SQUID (Software QUality In the Development process) approach, which combines many results from previous research on software quality and the European Commission funded projects SQUAD and CLARiFi. The application of MOSME is illustrated with a case study. We believe that a conceptual model of fully defined meaningful measures will help both the management board to give support to the data collection policy and the practitioner to avoid ambiguity in the definitions of the data measures.