Measuring the stability of data models

Abstract

The theory of data modelling makes a variety of claims about schema stability. The study reported in this paper developed a method whereby the major elements of a data model can be consistently represented whatever process was originally used for modelling. This was achieved through reverse-engineering a logical relational schema from the record design. The reverse-engineering process attempted to identify the primary meaningful primitives of a data model in order to track changes to them in different iterations of the application. The stability data collection process was applied to a case study followed by a series of models to generate further data. The early evidence indicated that data model instability has it roots in errors in modelling, errors in the semantic analysis whether done consciously or intuitively, and in changes to the requirements brought on by changes to the ‘reality’. This research suggested that some of the elements of a data model are significantly more important than others. The results of the analysis demonstrated that data models are not nearly as stable as the literature might indicate. The work suggested that there should be greater concentration on the question of data model evolvability, and on the appropriate preservation of meaning across model versions, rather than on data model stability.