Over the last weeks in Germany a broad discussion on the safety of nuclear power is taking place. I will not re-iterate this discussion, but I offer some help with the risk assessment… Many have argued that you cannot foresee major nuclear accidents – and I agree, but I disagree that you cannot estimate the risk. I think if you calculate the risk you may get worried...
What I learned from people who do risk assessment is very simple: past performance is an indicator of future performance. That means if you know how many car accidence have been caused by drivers aged 30-50 driving in one year (and what damage this caused) and if you assume that this number of accidents will happen in the following year then you are typically very close. And this is how insurances basically calculate the risk… (OK they use much more sophisticated models…, I recommend to read some books on insurance math - quite interesting).
What is now the risk of a catastrophic nuclear accident? Let’s doa rough estimate: There are about 500 nuclear power plants on the planet [1]. There are two catastrophic nuclear accident I remember (2011, 1986).
If we assume that each power plant has been running for 50 years or less, we have at most 25.000 “power plant years”.
Dividing this by 2 (as there were two catastrophic nuclear accidents I remember) the estimated risk is that a nuclear power plant will have a catastrophic nuclear accident about every 12.500 years (in fact the number should be lower as I a high estimate for the number of years running).
So for each power plant the risk that there is a catastrophic nuclear accidents in a single year 1/12.500 - which is very different from hitting the lottery jackpot…
For calculating the risk in Europe (which has somewhere in the order of 143 nuclear power plants [2]) within the next 10 years you can multiply the risk it by 143*10 and this is about 1/9. Obviously one could argue that this is not applicable to Europe because there were not accidents in Europe so far. But to take the analogy of the risks of drivers: if you do not know how women are driving – what is a better strategy to estimate the risk for women drivers: (a) assume they drive similar to men or (b) assume they do not have accidents at all?
Overall this calculation seems straightforward to me – but perhaps there is something fundamentally wrong with my estimate? Some people argue that small numbers of occurrences of events cannot be used to do such calculations... I do not really see why.
In general I think it is useful to know the estimated risks when making decisions, even if the estimates may be wrong by one or two orders of magnitute.
I wonder why insurances in Germany do not provide cover for y nuclear power plants [3].
[1] http://www.euronuclear.org/info/encyclopedia/n/nuclear-power-plant-world-wide.htm
[2] http://www.nytimes.com/2011/03/26/business/global/26nuke.html
[3] http://www.tagesspiegel.de/meinung/strahlen-und-haften/3968640.html
