Sam Keller's TEC Blog

Thursday, December 27, 2012

An Engineering Approach to Risk Analysis



Instead of reflecting on the unlikelihood of rare catastrophes after the fact, Elisabeth Paté-Cornell, a Stanford professor of management science and engineering and risk analysis expert , prescribes an engineering approach to anticipate them when possible, and to manage them when not.


Kelly Servick is a science-writing intern at the Stanford University School of Engineering.  In a recent article, she reviews the work of Elisabeth Paté-Cornell in this subject area. Click here for Ms. Servick's article.  Click here   for the link to Elisabeth Paté-Cornell's paper on the subject.

Ms. Paté-Cornell argues that a true 'black swan' - an event that is impossible to imagine because we've known nothing like it in the past - is extremely rare. (Reference "The Black Swan" by Nassim Nicholas Taleb.) The terms "black swan" and "perfect storm" have become part of the public vocabulary for describing disasters ranging from the 2008 meltdown in the financial sector to the terrorist attacks of Sept. 11, 2001. But using these terms too liberally in the aftermath of a disaster is really just an excuse for poor planning.

Her research on risk analysis was published in the November issue of the journal Risk Analysis.  Here she suggests that other fields could borrow risk analysis strategies from engineering to make better management decisions, even in the case of once-in-a-blue-moon events where statistics are scant, unreliable or even non-existent.

A true "black swan" – an event that is impossible to imagine because we've known nothing like it in the past – is extremely rare. The AIDS virus is an example. More often, there are important clues and warning signs of emerging hazards (e.g., a new flu virus) that can be monitored to guide quick risk management responses.
The 9/11 attack was not a black swan as the FBI knew that questionable people were taking flying lessons on large aircraft. 

Similarly, she argues that the risk of a "perfect storm," where multiple forces join to create a disaster greater than the sum of its parts, can be assessed in a systematic way before the event because even though their conjunctions are rare, the events that compose them – and all the myriad events that are dependent on them – have been observed in the past.



An engineering risk analysis is based upon systems, their functional components and their dependencies. For instance, many plants require cooling, generators, turbines, water pumps, safety valves and more all contributing to making the system work. Therefore, the risk analyst must first understand the ways in which the system works as a whole in order to identify how it could fail. The same methods can be applied to medical, financial or ecological systems.

Paté-Cornell says that a systematic approach is also relevant to human aspects of risk analysis.
"Some argue that in engineering you have hard data about hard systems and hard architectures, but as soon as you involve human beings, you cannot apply the same methods due to the uncertainties of human error. I do not believe this is true," she said.

In fact, she and her colleagues have long been incorporating "soft" elements into their systems analysis to calculate the probability of human error. They look at all the people with access to the system and factor in any available information about past behaviors, training and skills. by doing this, she has found that human errors, far from being unpredictable, are often rooted in the way an organization is managed. "We look at how the management has trained, informed and given incentives to people to do what they do and assign risk based on those assessments." 

Paté-Cornell has successfully applied this approach to the field of finance, where she has estimated the probability that an insurance company would fail given its age and its size. She has found that companies need forward-looking models that their financial analysts generally did not provide. Traditional financial analysis is based on evaluating existing statistical data about past events - like trying to drive by only looking in the rear view mirror.

In her view, analysts can better anticipate market failures – like the financial crisis that began in 2008 – by recognizing precursors and warning signs, and factoring them into a systemic probabilistic analysis.

Medical specialists must also make decisions in the face of limited statistical data, and Paté-Cornell says the same approach is useful for calculating patient risk. She used systems analysis to assess data about anesthesia accidents. Based on her results, she suggested retraining and recertification procedures for anesthesiologists to make their system safer.

"Lots of people don't like probability because they don't understand it," she said, "and they think if they don't have hard statistics, they cannot do a risk analysis." In fact, we generally do a system-based risk analysis because we do not have reliable statistics about the performance of the whole system.




 

No comments:

Post a Comment