This thesis is about psychological principles behind
conceptual modelling, when practised during the design of typical
business software application systems. The principles cover cognition
(perception, memory, mental models) and group dynamics. Our purpose
was to determine whether they could help make modellers more effective.
A critical analysis of existing methods was performed. A framework
of psychological principles was developed and was used to analyse
the design of an experimental modelling technique (method 'X'), consisting
of a graphical modelling technique and a software tool for visual
construction of models.
The
effectiveness of method 'X' in real business situations was compared
with that of more traditional object modelling. The results showed
stark differences in performance for untrained modellers between the
two techniques. Using object modelling, untrained modellers produced
results that were grossly incomplete and incorrect (22-35% on average).
Using the psychologically-inspired method 'X', untrained modellers
produced models that were almost complete and correct (universally
better than 82%). For an expert modeller quality was uniformly high
for both methods, but results indicate that expert modellers regarded
the quality of method 'X' models as better and preferred method 'X'
over the traditional approach. Significant differences in productivity
were observed at all levels of expertise. On average, the expert modeller
achieved productivity gains of approximately 150% with method 'X'
over object modelling, and untrained modellers achieved gains of over
450%. The results appear to support the idea of applying psychology
to conceptual modelling. They suggest that modelling need not require
a high degree of expertise, if methods and tools are adapted using
current psychological knowledge. The results could be exploited to
help untrained modellers, such as end users, who wish to develop or
participate in the development of large software systems but lack
access to the skills of trained IT professionals.