Network for risk professionals
What’s Wrong with Risk Matrices? Tony Cox's paper on RA journal, 2008
A couple of comments on the subject paper, a very interesting one indeed.
The author has been one of the few ones to point out in a rigorous way some serious drawbacks of risk matrices, drawbacks which many analysts sensed by intuition or lack of coherence in practical applications.
Most interestingly, Tony points out that when the risk matrix is used for intervention prioritization and consequence and probability are negatively correlated, the method turns out to be 'worse than useless'.
Now, in many fields it happens that P and C are just negatively correlated. such as industrial safety, where the number of fatalities and serious injuries, for example, is lower with respect to less serious injuries.
It also happens that in some countries, like mine (Italy) the use of risk matrices in industrial safety and hygiene is very common. I recently challenged a group of colleauges in a meeting, relating briefly Tony's arguments.
I received feeedback from two of 'em, one of 'em supporting Tony's contentions and saying that methods differnet from risk matrices should be followed like comparison with levels and tresholds reccomended by laws and regulations .
The other one affirmed that the use of risk matrices is anyhow useful to achieve a degree of uniformity in the classification of different risks, conceding that if regulations were adequately respected there would be little need for risk matrices.
Now, the issue is: is there an alternative quantitative way to assess different risks in working environments and achieve a reasonable degree of uniformity?
Such as: risk of noise, vibrations, chemical compounds, biological risk, cancer risk, muskoloskeletal disorders, various injuries, falls, fire and so on....
To me Bayesian analysis sounds promising. Here in Italy statistics about injuries are pretty much exhaustive. Such an analysis maybe should be implemented starting often by qualitative considerations.
Your comments and opinions are very much welcome.
Please find attached a plot and brief excerpts form Tony's paper (I hope my simple drawing of Tony's concept is correct).
The author has been one of the few ones to point out in a rigorous way some serious drawbacks of risk matrices, drawbacks which many analysts sensed by intuition or lack of coherence in practical applications.
Most interestingly, Tony points out that when the risk matrix is used for intervention prioritization and consequence and probability are negatively correlated, the method turns out to be 'worse than useless'.
Now, in many fields it happens that P and C are just negatively correlated. such as industrial safety, where the number of fatalities and serious injuries, for example, is lower with respect to less serious injuries.
It also happens that in some countries, like mine (Italy) the use of risk matrices in industrial safety and hygiene is very common. I recently challenged a group of colleauges in a meeting, relating briefly Tony's arguments.
I received feeedback from two of 'em, one of 'em supporting Tony's contentions and saying that methods differnet from risk matrices should be followed like comparison with levels and tresholds reccomended by laws and regulations .
The other one affirmed that the use of risk matrices is anyhow useful to achieve a degree of uniformity in the classification of different risks, conceding that if regulations were adequately respected there would be little need for risk matrices.
Now, the issue is: is there an alternative quantitative way to assess different risks in working environments and achieve a reasonable degree of uniformity?
Such as: risk of noise, vibrations, chemical compounds, biological risk, cancer risk, muskoloskeletal disorders, various injuries, falls, fire and so on....
To me Bayesian analysis sounds promising. Here in Italy statistics about injuries are pretty much exhaustive. Such an analysis maybe should be implemented starting often by qualitative considerations.
Your comments and opinions are very much welcome.
Please find attached a plot and brief excerpts form Tony's paper (I hope my simple drawing of Tony's concept is correct).
- Attachments:
-
-
Article_Cox_brief note.doc, 33 KB
-
Replies to This Discussion
-
Permalink Reply by William Huber on -
Luca,
I am very pleased to read you are interested in this issue. Tony and I are presenting new results at the SRA meeting next month (Boston, MA) in which we take a generous approach: given that people are going to continue using risk matrices, how can we design one to minimize the errors that might be made with it? We have established a pretty criterion for binary risk matrices (that is, risk is characterized into only two classes), the "zig-zag construction." Versions of it apply when you want to (a) reduce the worst mistake that can possibly be made using the matrix or (b) reduce the expected error, given a uniform probability distribution of cases. Situation (b) is amenable to Bayesian analysis, but in general (e.g., multimodal priors) the solution is difficult to obtain.
This seems to offer a middle ground between the points of view you have related.
I hope to have this paper ready for publication by the time of the meeting and plan to submit it to Risk Analysis for review afterwards.
-
Permalink Reply by Luca Nori on -
William,
All that sounds real interesting, is by chance a draft of your work available?
-
Permalink Reply by William Huber on
-
A draft will be available shortly before the SRA meeting. I hope to post a preliminary version on my web site (www.quantdec.com) under the "Articles" link.
About
© 2009 Created by William Huber on Ning. Create a Ning Network!
