 Can help this agency accept the recommendations of my favorite report. I'm all for it And I will follow it with a Greek dance not now The title of this speech over the years I have seen studies that take past incidents past accidents of nuclear reactors And through some statistical analysis They produce maybe the frequency of core damage or The frequency of releases, but it's really the core damage events that are usually of interest So I decided with some of my friends the staff whom I will name later to take a Look and See what is the issue there? What what do we learn from these analysis and what is the state of Of knowledge regarding the frequency of core damage Now Before I can think well, this is it I suppose Yeah, I think I need to explain what I mean by global statistical analysis We were we struggled with find trying to find a better terminology by that I mean the estimation of Core damage frequency or large early release frequency using accidents That happened at the high level at the plant level core damage events PRA Also uses statistics, but at a much lower level at the component level So that's the big difference that in a PRA we use analysis where identify Accident sequences and so on and then we use statistical information at the low level of components and human performance sometimes Whereas in the global analysis, it's at a very very high level so many core damage events worldwide Divide by the number of years and you get the estimate. I think I We it's important to understand How decisions are made and I believe the chairman alluded to this earlier today talking about The kinds of knowledge that we have to have when we make decisions Regulatory decision-making is based on what we know when we make the decision on The current state of knowledge It does not involve any Prophecies what prophecies as to what's going to happen in the future. It is what we know now And what is it that shapes this current state of knowledge? It's the design of the facilities it's the operation and and the regulations and We're informing the state of knowledge by science and the chairman talked about the earth sciences earlier this morning Engineering design and operation of the facilities and operating experience especially including past incidents So it's important to bear in mind That Probabilistic risk assessment does not predict anything. We are not trying to predict the future We are evaluating Possible evolutions of accidents based on what we know now we don't claim we are trying to predict what's going to happen and Enumerate future possibilities so that The decision-makers will have a better picture of what may happen and Make better decisions. So this is very important to bear in mind Nobody is trying to predict what will happen in the future and sometimes you hear, you know, did you predict three-mile island and so on? well There are thousands tens of thousands of sequences that are in a PRA One of them may predict what happened there will be thousands of others that didn't so these are Evaluations or assessments of possible evolutions of accidents now probability is not always an easy concept to use and There is a nice story here If you live to be a hundred rejoice Why? Because very few people die past that age Fundamental misunderstanding of the concept of conditional probability in the first statement and unconditional probability in the second Now that's not my joke it was told by well-known American comedian George Burns Now let's talk a little bit about how these global estimates remember at the plant level are produced Am I there? Yeah in order to be able to say something about the frequency of core damage of Current reactors and you want to use statistical analysis. You have to make sure that your statistical sample consists of reactors plants That are what is called in probability theory exchangeable events in other words is a current reactor of the same design of The same operation Procedures and under the same regulatory requirements as a reactor back in the same 70s back in the 80s Is it the same as a under the same operating under the same conditions as the Daiichi Plants in Japan were operating This is a fundamental assumption That we cannot ignore when we sample when we create a statistical sample So this exchange ability you will hear about throughout this speech So when people use this formula Dividing the number of incidents by the number of years There is this assumption behind it that the incidence at reactors That the incidence you are including your sample happened the reactors That are let's call them nominally identical with current reactors at least in this country and And it turns out that this is not quite true as we will see in the few in a few minutes Now there is a I think it's important to bear in mind this principle of exchange ability or assumption of exchange ability And I think there is another example that really brings the message home Okay, a professor Wilson of Harvard published a book a number of years back where he listed the riskiest professions in the United States and He stated that the riskiest professor a profession is being president That the probability of death of a president is 0.019 per year followed by fire fighters a factor of 48 lower and Police officers a factor of 59 lower Now he didn't comment on it But I thought that was strange So I tried to find out how that number was derived Well as most people in this audience know There have been four assassinations going back to 1865 So The number of years of the Republic Until the time that this calculation was done was 211 So if I use the formula I showed you a minute ago Divide 4 by 211 you get 0.019 So the question is then Is this really is 0.019 The risk of a modern president I Don't believe that I don't believe that it's riskier than Being a firefighter or a policeman So if you think about it now The answer is no Because what happened in 1865 1881 and so on The presidency the presidency's at that time are not exchangeable with the modern presidency The protections are Higher You might argue that the ways of attacking the president are also more sophisticated these days But the point is that the whole structure of the presidency and The protections that various federal agencies provide are very different from those of 1865 So this number is not valid It was derived under the false assumption of exchangeability now Even if we Accept that three mile island is not exchangeable with modern reactors It would be nice to know how to do the analysis, right? So if you take a statistical evidence one core damage event back in three mile island Over 38 or approximately 3800 reactor years of experience and you use What is known as Bayes theorem you get this uncertainty distribution Which ranges from something like eight? ten to the minus Ten to the minus is important for All the way down to ten to the minus six now This is the uncertainty that you would have derived if you had made the false assumption That three mile island is exchangeable with modern reactors So you might ask why do I say that? Well, look at the changes that have been and continue have been implemented and continue to be implemented after three mile island There was a large number According to some people a very large number of regulatory changes After three mile island and right now as he was pointed out by earlier speakers There are major changes in our regulatory structure because of Fukushima The Institute of nuclear power operations was established after three mile island that has had an impact on The way the facilities are operated The NRC established the individual plant Examination Program and the individual plant examination for external events program where the licensees identified vulnerabilities and they took action to fix them and a major Step forward is also the flex program that the industry is proposing and implementing Right now after Fukushima. So these are a few of the changes that really invalidate the assumption of exchangeability now what does what are the numbers that? Reasonably recent PRAs are telling us about core damage The I'm we have plotted here the point estimates that have been reported These are not uncertainty estimates unlike what I showed earlier the curve area was the uncertainty distribution This is just the point estimates of reactors as they have been reported and They range from ten to the minus four again all the way down to ten to the minus seven six the analysis is based on submissions by 61 units and 90% of these submissions Occurred after 2005 so they're fairly recent results if we if we move on to the large early release frequency The statistical analysis is even more problematic The reason being that there have been zero large leases in the United States And again if you take zero with 3800 Reactor years if you divide of course you get zero But if you go to base theorem you get a distribution that you see here and the numbers range from ten to the minus Four to ten to the minus nine a tremendous Range of uncertainty and all this says is that this range is consistent with seeing zero events in 3800 years now the The distribution of the Lurf estimates from PRAs again, it's this is based on 55 units 90% after 2005 It's a much it's a smaller range ten to the minus five all the way down to ten to the minus six and again The statistical analysis is not widely Inconsistent with the PRA estimates although again the statistical analysis is questionable so to reinforce well, first of all, I have already said that exchangeability doesn't apply but The question then is so all these accidents in the past are useless of Course not of course not. We are learning a lot from each Accident or incident and again this morning previous speakers talked about The changes that were implemented after Fukushima so This is a continuous learning process in other words the value of the experience with the accidents is the Qualitative insights what happened trying to understand why it happened. What can we do so that it will not happen again? so here is a list of Lessons learned and actions taken after three mile island A Lot of auxiliary Feedwater systems at the time before the accident were only manually actuated after the accident all of them converted to automatic actuation and There is a number of things here the emergency planning regulations were upgraded There were requirements related to hydrogen control Operator training was improved Stuffing requirements and the program on fitness of for duty was established If you look at Fukushima we As I was mentioned earlier the Commission has issued three orders that are being implemented right now so We required mitigation strategies for beyond design basis accidents We are studying very vigorously the issue of Multi-unit accidents although in the United States There are only a few sites that have three units It was mentioned earlier also about the severe accident capable and containment vents for BW ours with mark one and two Containments and Well, it's clear what we've done But it's important also to bear in mind that we are not learning only from experience from our operating experience Analysis is contributing a lot to our state of knowledge Going back to the mid 70s the first probabilistic risk assessment the reactor safety study Pointed out the importance of the small loss of coolant accident Because at that time everybody was focusing on the large loss of coolant accident the significance of human errors and support systems That study was followed by two industry sponsored PRAs in the early 80s those for Zion and Indian Point Which for the first time identified the significance of fires and earthquakes to Nuclear power plant risk people were aware of the seismic risk and fires and so on there was a fire protection program before that But really the focus was on the large break of the pipe And it was for the very first time interestingly enough by a study sponsored by the industry that these two major contributors to risk were identified and this finding has been confirmed time and time again in the last 30 years in The 80s we received the word from France That our colleagues there had done a PRA for low power and shutdown modes And they found that the risk from those modes was comparable To those to the risk from power operations That was a surprise to people at least in this country and because being shut down was considered a safe mode and That changed the way we looked at low power and shut down that did not come about Because of operating experience somebody sat down and did a PRA and identified it and The last point I want to make here is that another finding over the numerous PRA's that have been done over the years for plants Demonstrate that the risk is plant specific and a good example is again going back to the Indian Point studies of 30 years ago Where Indian Point 2 and 3 are called sister units yet the dominant contributors are different At that time the dominant contributor to latent Health effects was fire in one of the units and the other one it was earthquakes So what is the message here global statistical analysis requires the assumption? That at least if you want to do it for the United States that three mile island to is exchangeable with curry reactors It is not It is the qualitative insights from operational experience that are most useful Of course in regulatory decision-making, but also for the industry It's not the frequency of core damage and release That is derived from that experience And the PRA results represent the current design operation and regulation so Regulatory decision-making as I said earlier Must be based on the best current state of knowledge And that state of knowledge includes is refers to the design operation and regulation of the facilities It is informed by science engineering and operating experience including those past accidents and Unfortunately the need for the assumption of exchangeability between past present past and present reactors or future reactors Invalidates the global statistical estimates and another message that I would like you to live with Is that PRAs do not predict anything? They evaluate and assess potential accidents and areas So that better decisions will be made so Mark Twain said something that is consistent with what I just told you Facts are stubborn things, but statistics are pliable