Perception of Software Quality: Aesthetics and Expertise.

In both the educational and "real-world" practice of computer science, it is clear that aesthetics, expertise, and their interactions play important roles. However, these issues, taken together, have not, in the past, been the focus of empirical studies in computer science. In the current study, experts and novices were presented with a series of twenty "software development snapshots." A software development snapshot consisted of a brief description of a software engineering situation, statement of specific software development problem, and a partial program solution to the problem, each presented on one page. Unlike other types of stimuli which are often used in behavioral experiments in computer science, the snapshots were both short enough to be manageable in an experimental setting and rich enough to preserve some aspects of real software engineering situations. Four independent variables were included: the classicalness of both the problems and partial solutions, and surface structure features of the solutions were varied systematically among the snapshot items; interest ratings on the task items were also collected. Preference ratings for the items were the dependent variables. The task was administered to 46 computer science novices and 47 computer science experts. The novices were typically computer science sophomores who had completed three or four computer science courses. The experts were typically computer science seniors who had completed six or more computer science courses, including a data structures course. The ratings of the two groups of subjects were analyzed in two different ways. Analyses of covariance revealed that novices based their preference ratings primarily on characteristics of the presented solutions, as well as on the surface level features of the items. As the expertise level of the subjects increased, their focus shifted to problem features, as well as to the abstract elements in the task items. In order to explore patterns of preference among the subject responses, two Guttman-Lingoes Smallest Space non-metric factor analyses were also performed. For the experts, four categories of responses were found. These categories differed in terms of the scope and novelty of the problems and the appropriateness of the solutions for the problems. For the novices, five categories of responses were found. The novices' categories differed in terms of the familiarity and challenge of the solutions. The results of this study suggest that with increasing expertise, problem representations tend to become more abstract and more likely to incorporate problem features. This result is consistent with previous studies of expertise. Furthermore, preference categories for both subject groups indicate that both degree of structure and sustainability of the stimuli affect computer science preference.PhDComputer scienceUniversity of Michiganhttp://deepblue.lib.umich.edu/bitstream/2027.42/161424/1/8712162.pd