How could investigators use safety models to inform decisions on what to focus on?

The principal aim of air safety investigators is to analyse and understand systems to improve aviation safety by approaching each event as holistically as possible. To facilitate this, literature offers an overwhelming number of safety/accident models (SAMs), each of them reflecting a mixture of social, engineering and management perspectives. Although models are general in purpose, they suggest causality mechanisms (i.e., the possible “why” of events) and can drive the methods through which investigations are conducted (i.e., the necessary “how” to examine the possible “why”).Typically, researchers use SAMs to analyse investigation reports retrospectively by distilling causal and contributory factors, showing their relationships and interactions, presenting aggregated data from several reports, etc. Studies on individual SAMs have identified a spectrum of strengths and limitations, and research has also compared various SAMs (e.g., Swiss Cheese Model/HFACS, Accimap and STAMP). Still, these studies rely on the analysis of causal/contributing factors from published reports that are mapped against different models to compare similarities/differences in representations, difficulties in using those SAMs, etc.As such, little attention has been paid to how we can use SAMs proactively, before and during investigations, to inform decisions about where and what matters to look at in a system. During the presentation, we will suggest another, allegedly more useful, way to use the various SAMs. Our premise is that consultation with one or more SAMs can complement prior system knowledge and intuition from professional experience, stimulate discussions amongst investigation team members and help to make more deliberate decisions about what we should investigate or what we could presume as “true” and not necessitating investigation.Our presentation is based on the results from the analysis of two events from aviation: one incident and one accident. Based on the investigation reports, we distilled information about what the investigation team searched, regardless of whether this was found as causal/contributing to each event or not. Then we used four models (Swiss Cheese Model, AcciMap, STAMP and FRAM) to map this information. The mapping revealed what areas/elements from each model were searched by the investigators and the system areas suggested by these SAMs but not found in the reports.Expectedly, in the case of the Swiss Cheese Model and AcciMap, which are relatively less complex with fewer categories and system representation details, the “non-investigated” elements were fewer compared to STAMP and FRAM. Yet, the mapping revealed some system areas the investigators did not mention in their reports. With STAMP and FRAM, which include a higher detail of analysis, the “non-investigated” parts were more.Considering the resource limitations during investigations and the diverse backgrounds and knowledge of investigators, our position is that a similar conceptual mapping before and during investigations would be highly supportive. When planning investigations and/or after data collection begins, teams/investigators could start with consulting a “simple” SAM to identify system areas of interest they might have missed, and then, depending on constraints and other factors, extend this mapping by using more detailed SAMs. Indeed, we do not recommend that the models we used in our analysis should be the ones consulted by investigators. Also, we do not suggest that the mapping process should be something prescribed. Moreover, we acknowledge that despite various SAMs exist, some of them having been adapted by investigation authorities and agencies in their original or further modified versions, investigators do not (always) start their tasks with a model in mind. Nevertheless, SAMs can function as vehicles to improve investigation practice despite the weaknesses they inevitably carry.