Capture-Recapture Models to Estimate Software Fault Content
In order to control software inspections, it must be decided whether an inspected document is of sufficient quality and can be passed to the next development step. To make this decision the number of remaining defects in the document would be useful. Capture-recapture models, used in biology to estimate the size of animal populations, can be applied to inspections to estimate this number. Existing studies on the adoption of capture-recapture models for inspections focused mainly on two specific models and did not evaluate these models with real inspection data. This thesis presents a survey of existing types of capture-recapture models and identifies the capture-recapture models suited for inspections. These identified models are evaluated with respect to their accuracy (bias, variability, failure rate) using data gathered from actual inspections with real inspectors, real documents, and real defects. Factors that have an impact on accuracy and were considered in this thesis are the nu mber of inspectors involved in an inspection, the number of defects in an inspected document, and the reading technique inspectors use to analyse the document. The considered reading techniques are an adhoc-reading technique and a perspective-based reading technique. Important results of the evaluation are 1) that for a low number of inspectors the estimates are not very accurate, 2) that both adhoc-reading and perspective-based reading provide accurate estimates when using a sufficient number of inspectors, 3) that the accuracy depends strongly on the environment, and 4) that the models considered in the past for inspections are less optimal.
Kaiserslautern, Univ., Dipl.-Arb., 1997