Evaluation of using Capture-Recapture Methods on Software Review Data

Software defect content estimation is important, to control and ensure software quality. One possible method to achieve this is by applying capturerecapture methods. This type of methods can be applied on data collected from reviews, and it can be used to estimate the remaining number of defects after a review. This paper focuses on replicating a previous study made by other researchers. The replication is concerned with looking at different estimation methods and the ability to select a good enough estimation method. The evaluation in this paper is based on three data sets from three different types of documents: a text document, code and a requirements document. It is concluded that no estimation method is generally superior, and the best method is dependent on several factors, including the actual data sets and the number of reviewers. Moreover, it is concluded that the selection method proposed in the previous study is not superior to one specific model. Thus, the selection procedure is unable to pinpoint the best estimation method in our data sets.