1 . Complex , dynamic and open systems are difficult to understand and model . 2 . Infrequent events can have significant impact , but frequency data is not available . 3 . Estimating probabilities and severities can be subject to bias and inaccuracy , especially when choosing which outcomes and events to consider . 4 . Meaningful and pragmatic metrics can be hard to create . 5 . Tradeoffs are hard to make .
In dynamic and complex systems root cause analysis no longer suffices since many factors can contribute to outcomes 21 . “ Normal Accidents ” can be expected to occur 22 . Probabilities are often not meaningful or appropriate . Other approaches , such as using a systems model and using an understanding of losses , hazards , unsafe control actions and necessary constraints can offer an approach that can identify more problems , beyond component failure scenarios , find them earlier and at lower cost since they can be understood before design , development and implementation 23 , 24 , 25 .
It is hard to go from an understanding of risks to indicators that allow measurement and improvement . For a workplace safety example , safety metrics such as ‘ days since the last accident ’ look backward and correspond to the concept of ‘ free from unacceptable risk ’ and are not necessarily useful for “ safety in the future ”. They do not address chance , do not incorporate learning about the system and may give rise to complacency if there have not been any accidents over time 26 .
The use of probabilities in risk analysis can themselves raise questions since they can be hard to understand and establish . Randomness is variability that is fundamental while uncertainty can reflect a lack of knowledge and can be reduced . There can be uncertainty with the modeling , with lack of evidence to estimate probability distribution parameters , and with assumptions . Incompleteness includes “ known unknowns ” and “ unknown unknowns .” All of these uncertainties indicate that traditional risk management is not a complete solution and even with best effort incidents and accidents may still occur . This can be addressed with a systems model approach as well as resilience engineering .
21
Hollnagel , “ Resilience Engineering and the Future of Safety Management .”
22
Charles Perrow , Normal Accidents : Living with High-Risk Technologies , Princeton Paperbacks ( Princeton , N . J : Princeton University Press , 1999 ).
23
Leveson , Engineering a Safer World .
24
25
26
John Wreathall , “ Monitoring – A Critical Ability in Resilience Engineering ,” in Resilience Engineering in Practice : A Guidebook , ed . Erik Hollnagel et al ., Ashgate Studies in Resilience Engineering ( Farnham , Surrey , England ; Burlington , VT : Ashgate , 2011 ).
26 July 2022