Abstract
Re-inspection has been deployed in industry to improve the quality of software inspections. The number of remaining defects after inspection is an important factor affecting whether to re-inspect the document or not. Models based on Capture-Recapture (CR) sampling techniques have been proposed to estimate the number of defects remaining in the document after inspection. Several publications have studied the robustness of some of these models using software engineering data. Unfortunately, most of the existing studies did not examine the log linear models. Thus little is known about their robustness and applicability to software inspections' data. In this study we focus on the degree of accuracy and reliability of the log linear models with respect software inspection data. In order to explore the performance of the log linear models, we evaluated their performance for three person inspection teams. Furthermore, we evaluated the models using an inspection data set that was previously used to assess different CR models. Generally speaking, the study provided very promising results. According to our results, the log linear models proved to be more robust that all CR based models previously assessed for three-person inspections.