A Validation of Coupling and Cohesion Metrics for Object-oriented programs

Abstract

The emirical evaluation of software metrics is often questioned for lack of experimental rigor[8]. Most validations of previous software metrics were relied only on the analysis of correlation degree between metrics and software quality. The reliance, based solely on the correlation data, may be limited[4][8]. By combining the rigorous experiment framework [1] and the validation methodology [12], in the paper, we propose a validation approach specifically for coupling and cohesion metrics of object-oriented software. With four phases of an experiment: definition, planning, operation and interpretation [1], our approach can help structure experiment process and provide a classification scheme foe understanding and evaluating the experiment. In the planning phase, two kinds of direct reflectors for coupling and cohesion of object-oriented software respectively are proposed. A set of statistical models is provided for complete analysis between the direct reflectors and the metrics. With the analysis, the properties of quality assessment, quality control and quality prediction of a metric are evaluated properly. Following the proposed validation approach, we conducted an experiment to evaluate the effectiveness of several existing coupling and cohesion metrics of object-oriented software empirically. The experimental results indicate that message passing is the major contribution of coupling of object-oriented software.