An Empirical Investigation into Problem Decomposition Strategies used in Program Design

Abstract

In this thesis, findings are presented of a research investigation into general strategies for, and the effect of certain factors relating to, problem decomposition used in program design. The investigation involved two empirical studies, totalling six separate experiments, in which subjects trained in the broad principles of structured programming were asked to undertake various program design tasks associated with particular programming problems, solutions to which can be mapped through the use of "process structure hierarchies" onto a small number of "process decomposition paradigms." Analysis of the results revealed that solutions based on primitive, as opposed to abstract, perceptions of problem structure were strongly preferred, initially easier to perceive though harder to complete and were more error-prone. A model of program designer behaviour together with generalised problem decomposition strategies are advanced, that view program design as a problem-solving activity. These proposals form an explanatory framework for interpreting the experimental results, which are shown to be consistent with the proposals. In particular, it is argued that aspects of problem presentation and subject familiarity with component parts of a problem, are major factors that influence problem decomposition, and were responsible for the observed strong bias towards simplistic solutions. Additionally, it is argued that such bias can also be caused by "perception difficulty" allied to inadequacies in abstraction skills attributable to previous training. The thesis concludes with a recommendation that more specific, "criteria-driven" forms of structured programming need to be taught and practiced.