An investigation of human error in software development

Abstract

Context: Software defects occurring in code bases lead to an increased cost for software production and maintenance. To err is human nature and the process of software development is human centric. My analysis of the literature shows that the use of human error theory is emerging as an important tool for software development. Aim: The aim of my thesis is to present a training tool aimed at reducing the number of human errors developers make while working within the development phase of the Software Development Life cycle (SDLC) by improving developer situation awareness. Methods: My first study uses semi structured interviews to gain insight into what Skill-based (SB) errors developers make and how they mitigate these errors. My second study employs an experimental setup where developers log all human errors they make during developmental tasks across two weeks. At the beginning of week two the developers are asked to complete an online training package which I have developed on situation awareness. Results: The first study shows that the complexity of the development environment is one of the most frequently reported reasons for errors. I found that software developers struggle with effective mitigation strategies for their errors, reporting strategies largely based on improving their own willpower to concentrate better on development tasks. The results from the second study show that training software developers in situation awareness does lead to a decrease in the number of human errors made by those software developers. Conclusion: My doctoral research shows that human errors are a problem for software developers and loss of situation awareness is key for many of these developers. My preliminary results show that training tools which address situation awareness can aid developers in reducing the number of human errors that they make. Further work is required to investigate other means of improving developer situation awareness and determine whether my findings are generalisable.