Data Set: Software-Related Fatal Failures: An Empirical Exploration of RISKS Reports

About the dataset This dataset accompanies the study Software-Related Fatal Failures: An Empirical Exploration of RISKS Reports. It contains 73 documented cases, dated from 1978 to 2014, in which software was considered a crucial contributing factor to an accident involving one or more fatalities. The cases were identified primarily through the ACM SIGSOFT Software Engineering Notes series RISKS to the Public in Computers and Related Systems. The dataset also records supporting references such as official investigations, scientific publications, books, newspapers, and magazines where these were available. Purpose The dataset was created to provide an empirical basis for investigating software-related fatal failures. It supports research into: fatalities reported in connection with software-related failures; the probable main causes of these failures; the industries and application areas in which they occurred; the roles of software, hardware, users, interfaces, and operating conditions; and the quality and completeness of publicly available evidence. It is intended as a starting point for replication, validation, correction, extension, and comparative research in software safety, dependable systems, human-computer interaction, accident analysis, and risk management. Dataset contents Each row represents one reported case. The fields describe the date, number of deaths, location, nature of the accident, principal contributing faults, probable main-cause category, industry or application area, assessed data quality, references, and relevant notes. Probable main causes are classified as Physical, Software, Physical and Software, User-Software Interaction, or Insufficient Data. Evidence quality is classified as Good, Poor, or Controversial. Important limitations In this dataset, software-related does not mean that software was necessarily the sole or root cause. Many accidents involved interactions among software, hardware, users, interfaces, operating procedures, training, environmental conditions, and organizational factors. The collection is based on publicly available reporting and should not be interpreted as a complete census of software-related fatalities. Cases may be missing, incompletely investigated, disputed, or documented only through secondary sources. Fatality counts and causal classifications should therefore be treated as research estimates rather than definitive legal or technical findings. Reuse Researchers extending this dataset are encouraged to verify cases against primary or official sources, preserve the original case numbers, document all changes and reclassifications, and provide provenance information for newly added evidence. Suggested citation Sycofyllos, Nikolaos, and Eduard Paul Enoiu. 2021. Data Set: Software-Related Fatal Failures: An Empirical Exploration of RISKS Reports. Zenodo. DOI: 10.5281/zenodo.5493688.