The software inspection: The task and mechanisms for group support.

Abstract

Group software reviews play an important role in the software development process and may potentially benefit from group support. The purpose of this experimental study was to learn more about the nature of the software review task and to assess the impacts of selected group support mechanisms on review team performance. Two experiments were conducted. For each experiment, a review task involving the inspection of an analysis specification document (i.e., narrative, data flow diagrams, data dictionary) was performed by teams of undergraduate college student subjects. A review procedure based on Fagan's (1976) inspection technique was followed. To assess the impacts of group support on different aspects of the review process, the review task was split into two subtasks: defect identification and defect evaluation. Each subtask was investigated separately. The independent variable for the study was type of group support. The dependent variables were group performance and perceptions of the group process (e.g., sources of loss and gain). Group support was manipulated across treatments using non-electronic and electronic approaches. Electronic group support was provided by using an electronic meeting system. The experimental task was found to be a complementary task comprised of Eureka-type disjunctive subtasks. For the defect identification subtask, review teams receiving electronic process support (consisting of parallel communication and group memory) found significantly more defects than unsupported teams. The better performance of the electronically-supported teams was attributed to decreased process loss associated with domination and side-tracking. Satisfaction and acceptance of findings were rated higher for the unsupported teams compared to the electronically-supported teams. For the defect evaluation subtask, review teams receiving task support (consisting of a display of findings from the identification subtask) experienced significantly less process loss than unsupported teams. This outcome was attributed to the fact that the task support mechanism ensured that all findings from the identification session were raised to the review teams for evaluation. There were no differences with respect to satisfaction and acceptance of the findings. The implications of the study for researchers and practitioners are discussed.