Quality in data modelling: creating and understanding data models

This thesis aims to contribute to a theoretical understanding of factors that affect the creation and understanding of data models. Data models frequently form the basis of information systems analysis and design. Errors or omissions in the data model may lead to errors in the final system. It is considerably more cost effective to detect and correct errors in the early stages of development, such as in the data modelling stage, than it is to correct those same errors in the later stages of system construction.

This research concentrates on two specific elements the data modelling process, a) the creation of data models, and b) the understanding of data models.

This study employs a particular theory to investigate the creation of data models. It is argued that data can be abstracted or modelled in one of at least two ways: in terms of function or process, and in terms of data structure. This study proposes that it is the ability to abstract data structure from a natural language description, which is almost universally procedural in nature, that novice data modellers lack. It is this ability that educators should specifically target and aim for. An experiment was performed in a laboratory setting to test this proposition. It was found that when participants were required to perform structural abstraction, the quality of the model decreased.

This study also builds upon prior ontological theory that proposes that the proscription of optional properties from ER models will result in greater understanding of the model. However, prior experimental work in this area produced some puzzling results. This research proposes that short term memory was a mitigating factor in those prior experiments. The current study extends previous research by taking short term memory factors into account. One positive result was obtained. It was found that the proscription of optional properties lead to better delayed recall. To some extent, low statistical power may account for the other non -significant results in the current study.