Empirical Validation of Class Diagram Metrics.

Author(s): Marcela Genero, Mario Piattini and Coral Calero
Venue: IEEE: international Symposium on Empirical Software Engineering
Date: 2002


This paper investigates the usefulness of several established metrics for evaluation of class diagrams based on complexity. The metrics are analysed through two controlled experiment. The metrics evaluated in the experiment are: number of associations (NAssoc), number of aggregation (NAgg), number of dependencies (NDep), number of generalizations (NGen), number of aggregation hierarchies (NAggh), number of gernalization hierarchies (NGenH), maximum DIT (MaxDIT), maximum HAGG (MaxHAgg), number of classes (NC), number of attributes (NA), and number of methods (NM).

The authors set up two experiments in which they used computer science and software engineering students in their final year of study. The first experiment was set up to "validate the proposed measures as early maintainability indicators." The operation of the experiment consisted of 28 class diagrams in the domain Band Information Systems. Subejcts were then given all of the class diagrams along with questionnaire pertaining classes, attributes, associations, generalisations, etc. on their maintainability. The second study was a replica of the first to increase validity of the experiments.

The results of the experiments lead the authors to conclude that the metrics of NC, NA, NM, NAassoc, Nagg, NGen, MaxHAgg, and MaxDitt are to some extent correlated with maintainability in terms of understandability, analyzability, an modifiability. The authors used the Spearman correlation to support their conclusion.