Software Understandability: an Empirical Study (Comprehension, Requirements, Specifications, Expertise, Methodology).

The dramatic advances in computer hardware technology have been accompanied by intensified demands for understandable software. The understandability issue appears at every stage in the software life-cycle. Requirements specifications must be understandable to both user and designer. Numerous design techniques and programming language constructs have evolved in response to the understandability requirement. Maintenance programmers often must understand software planned, designed, coded, and used by others. Software must be easy to learn and easy to use; thus, understanding is a central theme. Although evaluation of the many software and notational tools and techniques is of central concern, understanding has proved difficult to measure. Many different types of definitions have been employed in efforts to measure it. It is necessarily a multi-faceted construct requiring multiple measures to encompass its complexity. The conceptualization and measurement of understanding are central to this dissertation. Cognitive theory provided a basis for the development of the measures while dimensional analysis was used to empirically validate the structure of the construct. An empirical investigation of the nature of understanding in the context of the human/computer interface was performed. Programmer and nonprogrammer understanding of requirements specifications written in structured English and Structured Analysis was studied experimentally. The results showed that understanding is, in fact, multidimensional. Further, specification language and programming expertise differed in their effects on understanding. In the case of low-level details of system operation, programming expertise had a significant effect using either specification language. For the more conceptual requirements of task-environment understanding, the advantage of programming expertise was evident only in the case of structured English. Programmers who read Structured Analysis were able to perform no better than nonprogrammers. In addition, the results for the subjective dimensions of understanding were not entirely parallel and provided a broader perspective on the effects of programming expertise and specification language on understanding.PhDComputer scienceUniversity of Michiganhttp://deepblue.lib.umich.edu/bitstream/2027.42/160500/1/8512387.pd