Investigating reinspection decision accuracy regarding product-quality and cost-benefit estimates

Abstract

After a software inspection the project manager has to decide whether a product has sufficient quality to pass on to the next software development stage or whether a second inspection cycle, a reinspection, is likely to sufficiently improve its quality. The reinspection decision of recent research focused on the estimation of product quality after inspection, which does not take in to account the effect of a reinspection. Thus we propose to use estimation models for the quality improvement during reinspection and the cost and benefit of a reinspection as basis for the reinspection decision. We evaluate the reinspection decision correctness of these models with time-stamped defect data from a large-scale controlled experiment on the inspection and reinspection of a software requirements document. The main finding of the investigation is that the product quality criterion is likely to force products to be reinspected, if a large number of defects were detected in the first inspection. Further the product-quality, criterion is especially sensitive to an underestimation of the number of defects in the product and will let bad products pass as good. The cost-benefit criterion is less sensitive to estimation error than the product-quality criterion and should in practice be used as second opinion, if a product-quality criterion is applied.