Requirements Defects over a Project Lifetime: An Empirical Analysis of Defect Data from a 5-Year Automotive Project at Bosch

Abstract

[Context and motivation] Requirements defects are notoriously costly. Analysing the defect data in a completed project may help to improve practice in follow up projects. [Question/Problem] The problem is to analyse the different kinds of requirements defects that may occur during the lifetime of an industrial project, and, for each kind of requirement defect, the respective number of occurrences and the cost incurred. [Principal ideas/results] In this paper, we present a post hoc analysis for an automotive project at Bosch. We have analysed 588 requirements defects reported during the elapsed project lifetime of 4.5 years. The analysis is based on a specific classification scheme for requirements defects which takes its eight attributes (incorrect, incomplete, etc.) from the IEEE 830 standard and refines them further by distinguishing nine possible defect sources (relating to parameters, wording, timing, etc.). The analysis yields that a large chunk of the requirements defects (61 %) stems from incorrectness or incompleteness. The requirements defects that are the most costly to fix are incompleteness and inconsistency. [Contribution] The insights gained from the analysis of the defects data allow us to review several design decisions for the requirements engineering process and to suggest new ones (such as to incorporate the classification of the requirements defects into the requirements review and into the defect reporting).