« The Nature of Creativity | Main | Into the Teeth of the Storm: The Blizzard of '78 »

December 21, 2009


Feed You can follow this conversation by subscribing to the comment feed for this post.

Rod Adams

The problem that I have with your analysis is the example that you have chosen. The US Nuclear Regulatory Commission recently spent about a half a decade updating risk models and performing detailed analysis about the consequences of even very bad accidents at nuclear power stations in the US.

The State of the Art Reactor Consequences Analysis (SORCA) project results showed that there is no chance (actually the probability was something like 10E -14) of even a single death from the kinds of accidents that might occur.


In other words, I agree that not all deaths are equal. I also agree that there is something even worse about a death when the person did not take any action to increase their own risk - like working in the industry.

That is one of the reasons that I spend so much time reminding people that natural gas, despite all marketing and advertising to the contrary, is not "clean" and not "safe". It releases many unseen contaminants right inside people's homes and into their water sources and it explodes with depressing frequency in cataclysmic situations like the one that occurred in San Bruno about 16 months ago.

Bob Kerns

Rod -- good comment (and I'm sorry it took me so long to approve it; I haven't been checking for comments as often as I should).

With no disrespect whatsoever to the people doing the SORCA analysis, I would argue that the odds that their calculations overlook some type of analysis to be far more likely than their result. To put it another way, risk analysis of unknown risks is an uncertain business, and we have to derate any such results on that basis.

So how do we do a meta-analysis, to estimate that risk? I have a few thoughts -- though far from rigorous.

One is that we look at the history of risk projections. Nuclear doesn't fare too well here, if we look at the *entire system*, including the governance and operations based on the analysis.

We can say the same about PG&E's risk analysis on pipeline testing. That's a great example -- because it shows, on a somewhat smaller scale, the sort of concentrated deaths I was associating with nuclear in my original article.

By the way, I picked nuclear for that side of the equation just because it something readers could relate to for that type of accident -- a low-frequency, very-high-cost event. Coal also has its distributed, low-profile events, as well as very dramatic mining accidents that devastate a community, in much the same way as your San Bruno example.

Except that in mining communities, these events are less of a surprise. In San Bruno, and Fukushima, there was an element of surprise -- an additional factor that affects how we deal with risks. Contrast, or example, tornadoes vs earthquakes.

The other avenue toward a meta-analysis I see, would be to look at it from complexity theory. How complex, how many steps, of how much uncertainty, are there in the risk analysis itself? Do these uncertainties compound? What external factors -- resource limitations, quality of data, political factors, confirmation bias on the part of the investigators -- impact the analysis?

I'm not saying we shouldn't try to do risk analysis! But for high-cost events, we should limit how much we trust negative results.

The comments to this entry are closed.