2we see them, is described in Section 5. Their implementation and all related practical issues arediscussed in Section 6.
Section 7 provides some insight into various strategies for dataanalysis. The interpretation of results is then described in Section 8. Finally, Section 9 identifiescommon types of measurement-based actions for improving the development process.
2. Motivation for Goal-Oriented Measurement
Measurement is introduced in software organizations to gain quantitative insight into thedevelopment processes and the developed products. This is important in order to understandbetter the development process, to identify problems and improvement opportunities.Measurement activities are commonly referred to as measurement programs.
A measurementplan specifies the why, what, how, and who of a measurement program.Goal-oriented measurement is the definition of a measurement program based on explicit andprecisely defined goals that states how measurement will be used. In addition, explicit modelshave to be defined to support the derivation of questions and measures from the goals in atraceable and unambiguous manner. Three main types of models may be required: descriptive,evaluation, and predictive models. For example, there may be a measurement goal dealing withproductivity. In this case, a
model is needed to define operationally whatproductivity is in the context of this measurement goal and environment, and what are theunderlying modeling assumptions. Another measurement goal may purport to determinewhether a component has a sufficient level of quality (e.g., based on a combined analysis of itscomplexity, coupling, and cohesion) to go into configuration management. In this case, an
model would be required. Finally, to provide an estimated value for a dependentvariable based on independent variables, e.g., project effort based on project size, teamexperience, and other influential project characteristics, one would need to specify and build a
model.Goal-oriented measurement helps
ensure adequacy, consistency and completeness of the measurement plan andtherefore of data collection. The designer(s) of a measurement program (referred toas measurement analysts) must deal with a large amount of information andnumerous interdependencies. In order to make sure that the set of measures isadequate, consistent, and complete, the measurement analysts need to know preciselywhy attributes are measured (e.g., size needs to be measured to predict project costearly in the process), what are the underlying assumptions (e.g., no code reuse), andin which models measures are intended to be used (e.g., regression model,COCOMO-like model).