Using subjective defect estimation models for reinspection decision

Abstract

To control software product quality, managers need accurate and timely feedback on the quality of the software product being developed. Inspection is one of the most effective means to detect defects early on and estimate the quality of the final product. Especially, in some quality critical cases, software documents or systems should be re-inspected to identify defects that have been missed during the initial inspection. The criteria for making the reinspection decisions are: $\it{the quality of the document and the quality of the inspection.}$ Most organizations have not institutionalized reinspection as a technique for ensuring the quality of the document and the inspection process. Most use historical experiences or follow a rule of thumb to determine whether to perform a reinspection [5]. The motivation of defect estimation models is ``providing better means for making the reinspection decision``. Two major, distinct defect estimate approaches are introduced: $\it{Objective models}$ based on refined team defect data, and $\it{subjective individual estimates}$, which are accessible during inspection with little extra effort. Subjective models can be easily integrated into existing inspection processes without extra effort to collect data or changing inspection process, while objective models require collecting data for statistical calculation, and changing inspection process etc. In addition, El Emam et al. and Biffl evaluated the accuracy of subjective models: They concluded that subjective estimations are equal or more accurate in industry situations. In this paper, we applied several subjective defect estimate models to the experiment and evaluated the accuracy of each model. The most accurate model``s estimate results were applied to the reinspection criteria. The validity of reinspection criteria was analyzed with the true results of the experiment. The main findings of the experiment were as follows: (1) Subjective defect estimation models t...