Expert Judgments about RD&D and the Future of Nuclear EnergyLaura D. Anadon,†,* Valentina Bosetti,‡ Matthew Bunn,† Michela Catenacci,‡ and Audrey Lee†,§

†John F. Kennedy School of Government, Harvard University, 79 John F. Kennedy Street, Cambridge, Massachusetts, 02138, UnitedStates‡Fondazione Eni Enrico Mattei, Corso Magenta 63, 20123, Milan, Italy§California Public Utilities Commission, 505 Van Ness Avenue, San Francisco, California, 94102, United States

*S Supporting Information

ABSTRACT: Probabilistic estimates of the cost and perform-ance of future nuclear energy systems under different scenariosof government research, development, and demonstration(RD&D) spending were obtained from 30 U.S. and 30European nuclear technology experts. We used a novelelicitation approach which combined individual and groupelicitation. With no change from current RD&D fundinglevels, experts on average expected current (Gen. III/III+)designs to be somewhat more expensive in 2030 than theywere in 2010, and they expected the next generation of designs(Gen. IV) to be more expensive still as of 2030. Projectedcosts of proposed small modular reactors (SMRs) were similarto those of Gen. IV systems. The experts almost unanimouslyrecommended large increases in government support for nuclear RD&D (generally 2−3 times current spending). The majorityexpected that such RD&D would have only a modest effect on cost, but would improve performance in other areas, such assafety, waste management, and uranium resource utilization. The U.S. and E.U. experts were in relative agreement regarding howgovernment RD&D funds should be allocated, placing particular focus on very high temperature reactors, sodium-cooled fastreactors, fuels and materials, and fuel cycle technologies.

■ INTRODUCTION

Nuclear power may prove to be one of the key technologies theworld uses to respond to climate change, but it faces manychallenges. Integrated assessment models of future energy andclimate paths vary widely in their projections of future nuclearenergy growth.1−3 Studies that place no constraints on nuclearenergy tend to project very large-scale growth, with nuclearenergy providing a significant fraction of future carbonreductions.3 But nuclear energy’s growth in recent years hasbeen very modest, with roughly four reactors per yearconnected to the grid worldwide on average in the pastdecade.4 Growth has been constrained by high costs and avariety of political, regulatory, and public acceptance challenges,which are likely to be exacerbated by the reaction to theFukushima accident in Japan. For nuclear power to displace abillion tons of carbon a year by 2050, roughly a tenth of what islikely to be needed to meet the internationally agreed goal oflimiting global average temperature increases to 2°C abovepreindustrial levels, would require adding 25 large nuclearplants to the grid every year from now until 2050.5 This meansnuclear energy would have to become much more attractive tothose making decisions about what types of power plants tobuild than it was in the decade before the Fukushima disaster.Development of improved nuclear technologies, offering

lower cost or improvements in areas such as safety, security,

proliferation-resistance, uranium resource utilization, and wastemanagement could address some of nuclear energy’s challenges,whereas others may be dependent on policy and politicalfactors. These nontechnological constraints may be affected byunpredictable events such as another large nuclear accident, aterrorist attack on a nuclear facility, or successes in siting andbuilding geologic repositories for nuclear waste.Decisions on research, development, and demonstration

(RD&D) investments, technology subsidies, and the like couldbe improved with better information on how the cost andperformance of key technologies might change in response tosuch investments. This paper provides detailed assessmentsfrom a range of U.S. and European technology experts of theexpected future cost and performance of three classes of nuclearreactor systems, large light-water reactors similar to thosecurrently available on the market (known as Gen. III/III+reactors), the next generation of designs (Gen. IV), and smallmodular reactors (SMRs) with sizes below 300 MWe. It alsopresents these experts’ judgments of how much governmentsshould spend on nuclear R&D, how those funds should be

Received: February 13, 2012Revised: August 29, 2012Accepted: September 24, 2012Published: September 24, 2012

Policy Analysis

pubs.acs.org/est

© 2012 American Chemical Society 11497 dx.doi.org/10.1021/es300612c | Environ. Sci. Technol. 2012, 46, 11497−11504

allocated, and what benefits such investments might bring inimproving nuclear energy cost and performance.Because the future of technology is inherently uncertain, it is

crucial to collect not just best estimates but judgments aboutthe likelihood of a range of outcomes. Such distributions cannotbe derived merely by looking at how a particular technology hasevolved in the past, or through simple learning-curve models.6

When past data is unavailable or of little use, the alternative isto rely on subjective probability judgments.7 For decades, manystudies have solicited experts’ subjective judgments of theprobability of uncertain events, for use as an input to thedecision-making process.8−10 However, both collecting esti-mates from experts and integrating these estimates into thedecision process present important challenges.11 As we discussin the next section, it is important to structure the elicitationinstrument to reduce overconfidence and other biases, therebyimproving the quality of the estimates.8,10,12−16

■ MATERIALS AND METHODSExpert elicitations are used to collect the views of expertsindividually, which means that experts do not have theopportunity to develop increased consensus through discussionamong the experts. Group-based methods such as the Delphiprocess17 can sometimes be unduly influenced or distorted bythe views of a small portion of the experts or by socialinteractions. To explore the potential value of government-funded research, development) investments in addressing thechallenges facing nuclear energy, we undertook a novel two-phase expert elicitation combining both individual interviews(conducted in an interactive online format) and a groupmeeting.There are several prior studies using expert elicitations on the

future of energy technologies,18−22 though our online protocolis unusual. Similarly, there are several studies that have usedstructured group-based elicitation methods,23,24 as well as casesin which a preliminary workshop has been organized before theindividual interviews to help survey design.25 But we are notaware of any instances in which individual elicitations werefollowed up by a group workshop, giving experts a chance todiscuss and revise their answers. We combined the twoapproaches because the literature suggests that individualelicitations are the most suitable approach for obtainingdetailed quantitative data from experts,14 as this method avoidsbiases stemming from group dynamics (e.g., social pressure26)and because conversely, group thinking and open discussioncan help experts move away from individual anchors andprejudices. The proposed combination of both methodsensured that each expert answered independently and, at thesame time, reacted to the stimulus of experts’ confrontation.The first phase of the elicitation consisted of a survey of 30

U.S. and 30 E.U. experts on nuclear technology during thesummer and fall of 2010, including a cross-section of expertsfrom private firms, government-sponsored laboratories, andacademia, with experience in several countries. Ten of the 30U.S. experts and four of the 30 E.U. experts were either in theprivate sector or had significant experience in the private sector.The experts included in this study have a wide range ofexperiences ranging from heads of nuclear units in theEuropean Union, consultants for international organizations,directors of research centers, university professors, and chieftechnical officers. Some have more experience in allocatingRD&D funding than others, but all are familiar with estimatingthe time and resources required to achieve a particular technical

goal. Of course, since RD&D funding decisions require trade-offs among many objectives, their recommendations on overallfunding should be considered one input among many. Giventhe uncertainties surrounding technical change, it is impossibleto know a priori which types of experts are more likely to becorrect about the impact of nuclear RD&D programs on futurecost reductions and improvements in noncost factors. A largebody of literature12−14 supports the inclusion of a mix ofexperts in expert elicitations to obtain a wide range of views onpossible technology futures, as it helps overcome the humantendency to anchor estimates to a single reference point.13

Resource limitations prevented us from broadening thesurvey to experts from Asia, Russia, and elsewhere, who shouldbe surveyed in future work. Participating experts spent 2−5 hcompleting an online interactive survey. Although the literatureon the use of expert judgment shows that most of theimprovement in eliminating common biases is already achievedwith as few as three experts,13 the large number of reactors andof dimensions to the problem we were investigating led us tochoose a much larger number of experts. Typically expertelicitations are conducted through face-to-face inter-views,14−16,19,21,28 but the online approach made it possibleto elicit judgments from a much larger number of expertswithin time and cost constraints.The individual online elicitation included sections with

background information on current U.S. and E.U. publicinvestments in nuclear RD&D; recent estimates of the currentand future cost of different types of reactors; guidance to helpexperts reduce bias and overconfidence; and self-rating ofexpertise, among other elements.8,10 The Supporting Informa-tion (SI) includes a list of the experts who participated in thesurvey and the workshop and their affiliations, links to thesurveys themselves, more detail on the research protocol andthe structure of the online elicitation (including the graphicalstrategies that were devised), and some evidence indicating thatmotivational biases did not play a large role in the experts’answers on budget allocation.Experts were asked (a) for their projections of costs and

performance in 2030 for the particular Gen. III/III+, Gen. IV,and SMR systems they expected to be “most commerciallyviable” at that time under different scenarios for governmentRD&D funding; (b) for their recommendations concerninghow much governments should spend on RD&D, and howthose funds should be allocated, by specific technology (e.g.,lead-cooled fast reactor, very-high temperature reactor, fuelcycle, fuels, and materials) and by level of technologydevelopment (i.e., basic research, applied research, experimentsand pilots, and commercial demonstration); (c) for theirestimates of the likely results of their recommended govern-ment RD&D investments; and (d) for their views on theimportance of particular factors that might constrain nuclearenergy growth that are not likely to be resolved by governmentRD&D, including the probability and impact of particularevents that could affect nuclear energy growth either positivelyor negatively. Wherever they were asked to make a projectionabout overnight capital cost, experts were also asked to provideuncertainty bounds. They were also asked to describe the mainhurdles that RD&D funds would seek to address.The Gen. III/III+ class of reactors are all light-water reactors

(LWRs). The Gen. IV concepts prioritized by the Gen. IVInternational Forum (which our elicitation focused on) cover awide range, including a very high-temperature gas-cooledreactor; a molten salt reactor; a supercritical water-cooled

Environmental Science & Technology Policy Analysis

dx.doi.org/10.1021/es300612c | Environ. Sci. Technol. 2012, 46, 11497−1150411498

reactor; and sodium-cooled, lead-cooled, and gas-cooled fast-neutron reactors. SMR concepts include both LWRs andconcepts in the Gen. IV classes. In each case, we asked expertsto estimate cost and performance for the system they expectedto be most commercially viable in 2030, and to specify whichparticular system they were referring to.The second part of this exercise consisted of a one-and-a-

half-day workshop that took place in April 2011 in Venice witha subset of 18 E.U. and U.S. nuclear experts, still spanningacademia, the private sector, and national or governmentlaboratories. The objectives of the workshop were to: (a)determine areas where consensus exists and, conversely, wherethe most important disputes and uncertainties lie; (b) test thevalidity of the information collected in the first stage; and (c)inquire about the possible reasons for differences of opinionamong experts and across the Atlantic. In the workshop, expertshad access to all the data gathered during the first phase, andthey had the chance to discuss and compare their answers,explore the issues in greater depth, and modify their answers.The workshop occurred in the weeks immediately following theFukushima accident in Japan, making it possible to assess howthis event changed the experts’ judgments.

■ RESULTSProjected Costs of Nuclear Power. In many markets,

high capital costs and associated financing difficulties are amongthe largest factors slowing the growth of nuclear power. Most ofthe experts we surveyed did not expect major breakthroughs inreducing the cost of nuclear power from any of the technologiescurrently in development.The cost of nuclear energy is dominated by the initial capital

costs of building the plants. We elicited estimates of theovernight capital cost, excluding owners’ costs and financingcosts. Actual overnight costs in the past have varied dependingon how well-managed the project was, but the industry hasbeen offering standard estimates of overnight costs of futureplants assuming the projects are well-managed, and it appearsthat the participating experts followed this approach. Expertswere asked to specify the RD&D objectives that would beaddressed with the funding devoted to their top four areas,forcing experts to think through what system componentswould benefit most from improvements. Some experts,however, may have based their holistic projections of systemcosts more on the history of nuclear costs and the currentperceptions of the nuclear community than on a systematicanalysis of how much different cost components could bereduced. For Gen. III/III+ reactors, Gen. IV reactors, andSMRs, both E.U. and U.S. experts generally estimated overnightcapital costs in 2030 greater than $3800/kW (costs areexpressed in 2010 U.S. dollars throughout this paper). As apoint of comparison, a recent MIT study estimated theovernight capital cost of new Gen. III/III+ nuclear power plantsin the United States at $4000/kW installed;29 most experts inour study offered generally similar estimates when assessingcurrent costs.In particular, if government RD&D spending continues at

the current rate of $466 million/yr in the U.S. and $800million/year in the E.U. (which we called a “business as usual”[BAU] funding scenario), half of the U.S. experts and 40% ofthe E.U. experts expected that the Gen. III/III+ designs wouldbe more expensive in 2030 than they are today (in constantdollars); 25% of the U.S. experts and 19% of the EU expertsthought costs would decrease modestly, with the remainder

projecting that costs would stay about the same (see Figure 1).Nearly all experts projected midrange Gen. III/III+ costs of

$3000−6000/kW in 2030 in a BAU scenario, with uncertaintyranges typically stretching from $2000−8000/kW. Since thesereactors are already designed, only a few experts projected thatexpanded government RD&D would reduce the cost of thesesystems substantially. Experts from the both sides of theAtlantic had similar estimates, although the most pessimisticbins (ratio of 2030 costs to today’s costs >1.2) are slightly morepopulated by U.S. experts. Figure S2 in the SI shows theexperts’ responses on the overnight capital cost of Gen. III/III+in 2030 under the BAU together with other RD&D fundingscenarios.While this expectation of a modest increase in cost may seem

surprising, the fact is that estimates of the capital costs ofnuclear power plants have been increasing steadily in recentyears. Recent studies have documented a “forgetting curve” inboth the United States and France (the two countries with thelargest numbers of deployed reactors), with costs rising as morereactors were built.30,31

Only five experts out of 60 foresee a cost for the mostcommercially viable Gen. IV system in 2030 lower than today’sGen. III/III+ reactors, assuming current public RD&Dexpenditures. Figure 2 shows the distribution of experts’ bestestimates for Gen. IV capital costs in 2030. Experts weredivided on which Gen. IV reactors would be most competitivein 2030. All of the Gen. IV designs except the supercritical-water-cooled reactor system were chosen by some experts, withthe two most frequently chosen concepts being the high-temperature or very-high-temperature reactor (HTR orVHTR) systems and the sodium-cooled fast reactor (SFR)system. Figure 2 shows the uncertainty range provided by eachexpert; only 14 experts (one-third of the total) estimated morethan a 10% chance of costs at or below a recent estimate from ateam from the U.S. nuclear laboratories.32

Whereas the participating E.U. experts uniformly expect thatthe cost of Gen. IV reactors in 2030 will be higher than the costof Gen. III/III+ reactors, the participating U.S. experts includedboth optimists and pessimists concerning the likely costs ofGen. IV reactors (see Figure 3). The high estimated costs ofGen. IV reactors in 2030 may be associated with the fact thatmost experts did not expect these systems to become

Figure 1. Distribution of experts’ ratio of estimate for Gen. III/III+capital costs in 2030 and 2010 under a BAU funding scenario.

Environmental Science & Technology Policy Analysis

dx.doi.org/10.1021/es300612c | Environ. Sci. Technol. 2012, 46, 11497−1150411499

commercially viable until later dates under a BAU fundingscenario, in some cases decades later (projected commercializa-tion dates are available in the SI, section S9). In particular, overhalf of all E.U. experts do not expect the main European Gen.IV project to be commercialized until after 2035, whereasalmost three-quarters of all U.S. experts expected some Gen. IVsystems to be fully commercialized by 2030 and to have lowcosts (through modular design and construction and inherentfeatures of some of the reactor systems); these factors mayexplain the transatlantic difference in 2030 cost projections forGen. IV systems.The participating experts were no more optimistic about the

costs of proposed SMRs with installed capacity below 300MWe. In a BAU public RD&D funding scenario, more thanthree-quarters of the U.S. and E.U. experts expected that themost commercially viable SMRs in 2030 would be moreexpensive in 2030 than Gen. III/III+ reactors, with almost two-thirds of the best estimates falling in the $4000−7000/kWrange. Experts provided overnight capital costs of SMRsbetween 100 and 300 MWe, based on discussions from theworkshop and comments made during the individual elicitation.

Several experts made clear that their SMR cost estimatesassumed a market large enough to allow mass-production in afactory. (Figure S3 in the SI shows the distribution of costanswers for SMR reactors.)In general, most experts offered fairly wide uncertainty

bounds on their projections of future costs (which is in linewith the poor record of past nuclear energy cost projections),and many experts offered ranges that were skewed upward, thatis, they estimated a significant chance that the cost might bemuch higher than their best estimate, but little chance that itwould be dramatically lower.In addition to the capital cost of building a nuclear plant, the

percentage rate that has to be paid to finance such a project isalso critical to the economic choice between building nuclearplants or other electricity sources. Both U.S. and E.U. expertsgenerally agreed that, given the various factors that might delayor block a nuclear plant, in 2010 a nuclear project would haveto pay investors a higher rateknown as a risk premiumcompared to, say, a comparable natural gas power project. Mostexperts expected this nuclear risk premium would decline by2030, but not to zero, meaning that in addition to high capitalcosts, nuclear energy would also suffer from a higher financingrate for those costs. More information on the financing rateresults is in the SI.

RD&D Recommendations and Impact on Costs andNon-Cost Factors. All but a few experts on both sides of theAtlantic recommended a large increase in government nuclearenergy RD&D. For both E.U. and U.S. experts, the fundinglevel recommended by the largest number of experts was 2.5−3times the BAU level (see Figure S5 in the SI).The participating experts generally agreed that their

recommended increases in RD&D would have a relativelylimited impact on future costs. Instead, expanded RD&D couldresult in improved performance in areas such as safety, wastemanagement, and uranium resource utilization, and could leadto new capabilities such as provision of high-temperatureprocess heat. Some reactor systems with desirable propertieswould simply not become available without additional publicRD&D investment. Both sets of experts agreed that beyondroughly $3 billion a year in the U.S. or in the E.U., increases innuclear RD&D investments would yield decreasing marginalreturns.Most experts indicated that increased public RD&D would

not change the cost of Gen. III/III+ reactors. With respect toGen. IV reactors, the participating E.U. experts projected thatexpanded RD&D would cut off the high tail of the projectedcost distribution. Half of the U.S. experts share this view, whilethe second half is fairly pessimistic about the effect ofexpanding R&D effects on costs. Only a few experts in eithergroup, however, projected that Gen. IV systems could reachcosts below $3800/kW installed by 2030 even under theirrecommended RD&D funding. Similarly, all but a few expertsthought that SMRs would cost $4000/kW or more in 2030,even under their recommended RD&D funding.We also evaluated the relationship between the 50th

percentiles of cost and the sector of the experts. Withoutcontrolling for any other factors, we found that industry expertswere more pessimistic than experts in public institutions (onaverage their cost estimates were 458 $/kW greater, with a p-value of 0.06) and that academics were more optimistic thanexperts in public institutions (on average their cost estimateswere 900 $/kW lower, with a p-value of 0.00).

Figure 2. Experts’ best estimates (black circles), and 90th and 10thpercentile error bands (vertical lines) for Gen. IV capital costs in 2030under a BAU RD&D funding scenario. The range of advanced fuelcycle (AFC) cost basis estimates is also provided for comparison(horizontal band in gray, the gray dotted line represents their bestestimate).32

Figure 3. Distribution of experts’ ratio of estimate for Gen. IV capitalcosts in 2030 over estimates for Gen. III/III+ in 2010 under a BAUfunding scenario.

Environmental Science & Technology Policy Analysis

dx.doi.org/10.1021/es300612c | Environ. Sci. Technol. 2012, 46, 11497−1150411500

Combining the answers of U.S. and E.U. experts regardingthe 2030 costs under increasing RD&D scenarios, we are ableto estimate the cost-reduction return from RD&D investments(see SI, section S7 for more details). With increasing publicRD&D investments over the BAU RD&D level, theparticipating experts expect small cost reductions (particularlyfor already-designed large-scale Gen. III/III+ systems) anddecreasing marginal returns per RD&D dollar spent. We usethe concept of a returns to RD&D curve (on the analogy to themore commonly used learning curve based on cumulativecapacity) drawing from the literature supporting the idea ofdecreasing returns to knowledge33−37 to extract a relationshipbetween RD&D investments under scenario i (RDi) andresulting 2030 overnight capital costs improvements over theBAU case. In particular, if c(RDBAU) are costs under the BAURD&D scenario, then the ratio of costs under scenario i tocosts under BAU can be defined as follows:

=β⎛

⎝⎜⎞⎠⎟

cc

(RD )(RD )

RDRD

i i

BAU BAU (1)

Where β is the returns to RD&D coefficient, implying that adoubling of RD&D would reduce costs by a fraction equal to1−2β.Table S2 in the SI shows that a best fit to the participating

experts’ projections implies that in the case of Gen. III/III+reactors, each doubling of RD&D investments would lead to anadditional cost reduction of only 1.36%. The correspondingfigure for Gen. IV systems is 5.13%, implying a larger, thoughstill modest, role for increased government RD&D in reducing2030 costs. SMRs would benefit similarly from additionalpublic RD&D, with a return to RD&D rate of 5.11%. The lowvalues of the R2 in the regression indicate that levels of RD&Dinvestment are not the only factor affecting projections offuture costs. An ANOVA analysis of the relationship betweenRD&D investments and the cost of the three reactor typesindicated that including an “expert” variable increases theexplanatory power of our specification, but does not affect thesign, value, or statistical significance of the RD&D component.

This concept of an exponential learning-by-researching curvewith decreasing marginal returns is most appropriate formodeling incremental improvements in the costs of anestablished set of technologies. We would argue this is agood description of nuclear energy, where most of the currentefforts are focused on improvements to reactor and fuel cycleconcepts that originated decades ago. But in addition toincremental improvements, increasing RD&D also increases theprobability of disruptive innovations that could lead to step-function shifts in cost or performance; these are more likely tooccur in some other technological areas than they are in nuclearpower at its current stage of development.As noted above, RD&D has many other purposes that go

well beyond reducing nuclear energy costs. The Gen. IVInternational Forumthe international consortium pursuingRD&D on the Gen. IV conceptshas defined a range of goals,and we asked the E.U. and U.S. experts to estimate how mucheach Gen. IV goal would be addressed by 2030 assuming BAURD&D funding. The scale ranges over five steps: the inside stepwould indicate that the goal would be fully addressed (the innerring of the pie chart in Figure 4); the outside step wouldindicate that the goal would not be addressed at all (the outerring of the pie chart in Figure 4). We used a Likert-scaleframing for these questions because they are preferable to yes/no questions38 and and because it allowed us to keep amanageable survey length. To maximize the quality of theanswers, we defined quantitatively what was meant by each ofthe five scales (see SI for passwords to the surveys to read thedefinitions). The numbers in each section of the rings are thecount of experts who gave that rating for that goal; the shadingsreflect this count.Both U.S. and E.U. experts were optimistic that a BAU

RD&D program would make major progress in addressingoperational safety and reliability. Both groups are fairlypessimistic about a BAU program addressing waste minimiza-tion and management, resource utilization, or proliferationresistance and physical protection issues, though E.U. expertswere less pessimistic than their U.S. counterparts (Figure 4).For both sets of experts, half of the ratings for life cycle costs

Figure 4. How much will RD&D programs under a BAU funding scenario contribute to addressing the different goals of RD&D programs (namely:resource utilization, waste minimization and management, lifecycle cost, risk to capital, operational safety and reliability, core damage, offsiteemergency response, and proliferation resistance and physical protection) by 2030 according to U.S. (left, Figure 4a) and E.U. (right, Figure 4b)experts? The innermost ring = fully address; second ring = significantly address; third ring = moderately address; fourth ring = only partially address;outermost (fifth) ring = not address. The color is proportional to the number of experts in that category: a darker color means that many expertsthought that BAU RD&D programs would have a particular level of impact in that particular goal. The graph refers to the Gen. IV reactor type thatexperts thought would be most commercially viable in 2030, which differs by expert.

Environmental Science & Technology Policy Analysis

dx.doi.org/10.1021/es300612c | Environ. Sci. Technol. 2012, 46, 11497−1150411501

are in the outer two rings (not addressed at all or onlyaddressed in part).Under their recommended budgets, 50% of the E.U. experts

foresee that all objectives will be at least significantly addressed(the question in the U.S. survey is not comparable). It isnoticeable that life-cycle cost and risk to capital are the onlytwo goals that are almost insensitive to the increase in RD&D.Even though these improvements by 2030 seem modest,experts thought that these benefits justified large increases inRD&D because many Gen. IV designs are more likely to play amajor role in the longer term (around 2050) and because of theimportant role that nuclear power could play in a world seriousabout reducing climate emissions. In the survey, the degree ofprogress was rated only as it affected likely future growth ofnuclear energy; some experts may have believed that progressthat would not greatly affect future growth but could reduce therisks of that growth would justify significant investment.Recommended RD&D Allocations. What particular

technological areas should an expanded nuclear RD&Dinvestment focus on? On average, both U.S. and E.U. expertsrecommend devoting roughly a quarter of the investment tofuel cycle technologies and fuel materials, and agreement acrossexperts on this point was relatively high (coefficients ofvariation between experts below 0.6) (Figure 5). Experts

disagreed more over which particular reactor systems deserveexpanded RD&D funding. Sodium-cooled fast reactors (SFR)attracted the largest average share of the experts’ recommendedbudgets, both in the U.S. and the E.U. surveys, followed by thevery high temperature reactor (VHTR).SFRs, like most of the Gen. IV systems, are based on

recycling spent fuel to extend uranium resources and, in somecases, improve waste management by transmuting long-livedisotopes. They have been under development since the 1950s ata cost of tens of billions of dollars in RD&D, but only a fewcountries expect to commercialize them before 2035.39

Although the SFR and fuel cycle technologies received thehighest average budget allocations, the rationale for focusing onfast reactors and recycling attracted the most polarizeddisagreement, with some experts arguing that uranium wascheap and abundant and any waste management improvementswere not likely to be substantial, and others arguing thatrecycling could offer decisive waste management advantages, orthat providing sufficient fuel to sustain large-scale growth ofnuclear power over an extended period would require recycling.The VHTR offers the potential to produce high-temperature

process heat for chemical and industrial purposes; this wouldincrease the reactors’ overall efficiency and make it possible toproduce electricity when electricity prices were high and otherproducts at other times. E.U. experts on average allocated lessto SMRs than their American counterparts, and in theworkshop, most E.U. experts indicated that they did notenvision any substantial market for smaller reactor systems.Experts were also asked to provide the rationales for their top

four funding allocations. For a summary of the top RD&Dobjectives identified by the experts as the main priorities forRD&D funding, the reader is referred to section S9 in the SI.Our analysis indicated that there was little correlation betweenexperts’ self-described areas of expertise and their RD&Dfunding allocations, suggesting little or no bias towardrecommending RD&D that would benefit their own projects(see section S9 in the SI for additional discussion).Sections S11 and S12 in the SI include a discussion of the

results on the constraints on the future growth of nuclearpower, and a discussion on the impact of the Fukushimaaccident on the estimates of the experts on the future growth ofnuclear power, respectively. The experts who participated in theworkshop made few changes in their funding allocations andonly modest changes in their projections of nuclear powergrowth in response to the Fukushima accident.

■ DISCUSSIONOur expert elicitation revealed broad consensus among a widerange of industry, academic, and government laboratory expertsin both the United States and the European Union that a largeincrease in government investment in nuclear energy RD&D isneeded, but will lead to only modest improvements in nuclearenergy costs. The primary benefits the experts envision are inother areas, such as improved safety and waste management oroffering new capabilities such efficient spent fuel recycling orprovision of high-temperature process heat. Different Gen. IVreactor designs offer possible benefits and trade-offs alongdifferent Gen. IV goals, which is why it may be too early tofocus on a single reactor design (see SI, section S8).Nuclear energy appears likely to continue on a path of slow

to moderate growth after the Fukushima disaster, though somecountries have turned away from this energy source. But ourresults raise serious questions about whether nuclear energy canachieve the dramatic growth required for it to play a significantpart in mitigating climate change, dealing with limited suppliesof liquid fossil fuels, or providing energy access to all. If theexperts’ cost projections prove to be correct, nuclear plants arenot likely to be competitive with coal and natural gas plants inmany major markets unless policies are put in place that havethe effect of creating a substantial price on carbon. RD&Dinvestments alone are not likely to be enough to achieve rapidnuclear energy growth. Such investments would have to becoupled with a range of government policies to support nuclearenergy, and with widespread industry and public support.

Figure 5. Average percentage budget allocation for E.U. (top, panel a)and U.S. (bottom, panel b) experts. The mean average allocation iscolor-coded from 0% to 15% (the redder the cell, the greater theaverage allocation). The coefficient of variation is color-coded from 0to 2.1 (the darker the gray, the greater the disagreement across expertsabout the budget allocation for a particular RD&D area).

Environmental Science & Technology Policy Analysis

dx.doi.org/10.1021/es300612c | Environ. Sci. Technol. 2012, 46, 11497−1150411502

Whether it will be possible to generate the broad supportrequired in the aftermath of the Fukushima disaster remains tobe seen.

■ ASSOCIATED CONTENT*S Supporting InformationThe Supporting Information includes the following sections:(S1) the list of experts participating in the individual onlineelicitations and in the group discussion; (S2) logins to the E.U.and U.S. online elicitation tools; (S3) information on thedesign of the elicitation instrument; (S4) a brief analysis ofcognitive and motivational biases; (S5) a discussion of insightsobtained from the two-stage methodology of the elicitation;(S6) figures with experts estimates of the overnight capital costsof large-scale Gen. IV and SMR reactors in 2030; (S7) resultson the returns to RD&D; (S8) a figure with the experts’recommended level of public investments in nuclear RD&D;(S9) a discussion of RD&D objectives; (S10) the expertsestimates about the risk premium of nuclear power plants overthose of natural gas plants at different points in time; (S11)expert judgments about nontechnical factors that may constrainthe future of nuclear growth; and (S12) the results and adiscussion of the impact of Fukushima on the expert estimates.This material is available free of charge via the Internet athttp://pubs.acs.org.

■ AUTHOR INFORMATIONCorresponding Author*Phone: (+1) 617-384-7325; fax: (+1) 617-495-8963; e-mail:laura_diaz_anadon@harvard.edu.NotesThe authors declare no competing financial interest.

■ ACKNOWLEDGMENTSWe thank the experts listed in Table S1 in the SupportingInformation for their assistance with this project, twoanonymous reviewers for their helpful comments, and JacopoCrimi for his work with the TOC art. The work was supportedby the Climate Change Initiative of the Doris Duke CharitableFoundation and the European Research Council under theEuropean Community’s Seventh Framework Programme(FP7/2007-2013)/ERC grant agreement n° 240895 − projectICARUS “Innovation for Climate Change Mitigation: a Studyof Energy R&D, its Uncertain Effectiveness and Spillovers”.

■ REFERENCES(1) Clarke, L.; Edmonds, J.; Krey, V.; Richels, R.; Rose, S.; Tavoni,M. International climate policy architectures: Overview of the EMF 22International Scenarios. Energy Economics 2009, 31, S64−S81Database..(2) Edenhofer, O.; Carraro, C.; Hourcade, J.-C.; Neuhoff, K.;Luderer, G.; Flachsland, C.; Jakob, M.; Popp, A.; Steckel, J.;Strohschein, J.; Bauer, N.; Brunner, S.; Leimbach, M.; Lotze-Campen, H.; Bosetti, V.; de Cian, E.; Tavoni, M; Sassi, O.;Waisman, H.; Crassous-Doerfler, R.; Monjon, S.; Droge, S.; vanEssen, H.; del Río, P.; Turk, A. RECIPEThe Economics ofDecarbonization. Report on Energy and Climate Policy in Europe.Potsdam Institute for Climate Impact Research, Centro Euro-Mediterraneo per I Cambiamenti Climatici, Centre International deRecherche sur l’Environnement et le Developpement, and ElectricityPolicy Research Group, 2009.(3) Goldston, R. J. Climate change, nuclear energy and nuclearproliferation: Magnitude matters. Sci. Global Secur. 2011, 19 (2), 130−165.

(4) IAEA. Power Reactor Information System; International AtomicEnergy Agency: Vienna, Austria, 2011.(5) Bunn, M.; Malin, M. B. Enabling a nuclear revivalandmanaging its risks. Innovations 2009, 4 (4), 173−191.(6) Grubler, A.; Nakicenovic, N.; Victor, D. G. Dynamics of energytechnologies and global change. Energy Policy 1999, 27, 247−280.(7) National Research Council. Prospective Evaluation of AppliedEnergy Research and Development at Doe (Phase Two); The NationalAcademies Press: Washington DC, 2007.(8) Morgan, M. G.; Henrion, M. Uncertainty: A Guide to Dealing withUncertainty in Quantitative Risk and Policy Analysis; CambridgeUniversity Press: New York, 1990.(9) Cooke, R. Experts in Uncertainty: Opinion and SubjectiveProbability in Science; Oxford University Press: New York, UnitedStates, 1991.(10) O’Hagan, A.; Buck, C. E.; Daneshkhan, A.; Eiser, J. R.;Garthwaite, P. H.; Jenkinson, D. J.; Oakey, J. E.; Rakow, T. UncertainJudgments: Eliciting Experts Probabilities. John Wiley and Sons, Ltd.:Chichester, U.K., 2006.(11) Keith, D. W. When is it appropriate to combine expertjudgments? Clim. Change 1996, 33, 139−144.(12) Clemen, R. T.; Reilly, T. Making Hard Decisions with DecisionTools, Duxbury Thomson Learning; Pacific Grove, CA 2001.(13) Keeney, R. L.; von Winterfeldt, D. Eliciting probabilities fromexperts in complex technical problems. IEEE Trans. Eng. Manage.1991, 38, 191−201.(14) Meyer, M. A.; Booker, J. M. Eliciting and Analyzing ExpertJudgment: A Practical Guide; Academic Press Ldt: London, U.K., 1991.(15) Phillips, L. D. Group elicitation of probability distributions: Aremany heads better than one? In Decision Science and Technology:Reactions on the Contributions of Ward Edwards; Shanteau, J., Mellers,B., Schum, D. A., Eds.; Kluwer Academic Publishers: Norwell, MA,1999.(16) Walls, L.; Quigley, J. Building prior distributions to supportBayesian reliability growth modelling using expert judgement. Reliab.Eng. Syst. Saf. 2001, 74, 117−128.(17) Dalkey, N. C. The Delphi Method: An Experimental Study ofGroup Opinion, RAND Report #RM-5888-PR; Rand Co.: SantaMonica, CA, 1969.(18) Baker, E.; Chon, H.; Keisler, J. Advanced Nuclear Power:Combining Economic Analysis with Expert Elicitations to InformClimate Policy, 2008, SSRN-id1407048(19) Baker, E.; Chon, H.; Keisler, J. Carbon capture and storage:Combining expert elicitations to inform climate policy. Clim. Change2009, 96 (3), 379−408.(20) Baker, E.; Keisler, J. M. Cellulosic biofuels: Expert views onprospects for advancement. Energy 2011, 36, 595−605.(21) Curtright, A. E.; Morgan, M. G.; Keith, D. W. Expertassessments of future photovoltaic technologies. Environ. Sci. Technol.2008, 42, 9031−9039.(22) Chan, G.; Anadon, L. D.; Chan, M.; Lee, A. Expert elicitation ofcost, performance, and RD&D budgets for coal power with CCS.Energy Procedia 2011, 4, 2685−2692.(23) IPCC. Fourth Assessment Report; Cambridge University Press:Cambridge, United Kingdom and New York, United States, 2007.(24) IPCC. Special Report on Renewable Energy Sources and ClimateChange Mitigation; Edenhofer, O., Pichs-Madruga, R., Sokona, Y.,Seyboth,K., Matschoss, K., Kadner, S., Zwickel, T., Eickemeier, P.,Hansen, G., Schloemer, S., von Stechow, C., Eds.; CambridgeUniversity Press: Cambridge, 2011.(25) Roman, H. A.; Walker, K. D.; Walsh, T. L.; Conner, L.;Richmond, H. M.; Hubbell, B. J.; Kinney, P. L. Expert judgmentassessment of the mortality impact of changes in ambient fineparticulate matter in the U.S. Environ. Sci. Technol. 2008, 42 (7),2268−2274.(26) Burman, M. A. Environmental Risk and Decision Analysis: ForConservation and Natural Resource Management; Cambridge UniversityPress: London, U.K., 2005.

Environmental Science & Technology Policy Analysis

dx.doi.org/10.1021/es300612c | Environ. Sci. Technol. 2012, 46, 11497−1150411503

(27) Hora, S. C. Probability judgments for continuous quantities.Manage. Sci. 2004, 50 (5), 597−604.(28) Zickfeld, K.; Morgan, M. G.; Frame, D. J.; Keith, D. W. Expertjudgments about transient climate response to alternative futuretrajectories of radiative forcing. Proce. Natl. Acad. Sci. 2010, 107,12451−12456.(29) Du, Y.; Parsons, J. E. Update on the Cost of Nuclear Power,Report 09-004; MIT Center for Energy and Environmental PolicyResearch: Cambridge, MA, May 2009.(30) Koomey, J.; Hultman, E. A reactor level analysis of busbar costsfor U.S. nuclear plants, 1979−2005. Energy Policy 2007, 35, 5630−5642.(31) Grubler, A. The costs of the French nuclear scale-up: A case ofnegative learning by doing. Energy Policy 2011, 38 (9), 5174−5188.(32) Shropshire, D. E.; Williams, K. A.; Hoffman, E. A.; Smith, J. D.;Hebditch, D. J.; Jacobson, J. J.; Morton, J. D.; Phillips, A. M.; Taylor, J.P. Advanced Fuel Cycle economic analysis of symbiotic light-water reactorand fast burner reactor systems. Technical Report from the Idaho NationalLaboratory, ocument INL/EXT-09-15254; U.S. Department ofEnergy: Washington DC, January, 2009.(33) Evenson, R. E. International invention: Implications fortechnology market analysis . In R&D, Patents and ProductivityZvi, G.,Ed.; University of Chicago Press: Chicago, United States, 1984.(34) Evenson, R. E.; Kisleve, Y. A stochastic model for appliedresearch. J. Political Econ. 1976, 84, 265−281.(35) Kortum, S. Research, patenting, and technological change.Econometrica 1997, 65 (6), 1389−1419.(36) Segerstrom, P. S. Endogenous Growth without Scale Effects.American Economic Review 1998, 88 (5), 1290−1310.(37) Bosch, M.; Lederman, D.; Maloney; W. F. Patenting andResearch and Development: A Global View; The World Bank:Washington DC, 2005.(38) Plous, S. The Psychology of Judgment and Decision Making;McGraw-Hill: New York, 1993.(39) Cochran, T. B.; Feiveson, H. A.; Patterson, W.; Pshakin, G.;Ramana, M. V.; Schneider, M.; Suzuki, T.; von Hippel, F. Fast BreederReactor Programs: History and Status, Research Report #8. Interna-tional Panel on Fissile Materials. Princeton University’s Program onScience and Global Security. Princeton, NJ, February, 2010.

Environmental Science & Technology Policy Analysis

dx.doi.org/10.1021/es300612c | Environ. Sci. Technol. 2012, 46, 11497−1150411504

  S1

Supporting Information (SI)

Expert Judgments about RD&D and the Future of Nuclear Energy

Laura D. Anadona *, Valentina Bosettib, Matthew Bunna, Michela Catenaccib, Audrey Leea,c

aHarvard University, John F. Kennedy School of Government, 79 John F. Kennedy Street, Cambridge, Massachusetts, 02138, United States bFondazione Eni Enrico Mattei, Corso Magenta 63, 20123, Milan, Italy cCalifornia Public Utilities Commission, 505 Van Ness Avenue, San Francisco, California, 94102, United States *Corresponding author: laura_diaz_anadon@harvard.edu, tel.: (+1) 617-384-7325, fax: (+1) 617-495-8963 This Supplementary Information document contains fifteen pages (including this cover sheet), nine figures (Figures S1-S9), two tables (Tables S1 and S2), references, and is divided into the following sections:

SI Section Page S1. List of the 60 E.U. and U.S. Nuclear Experts Who Participated in this Study S2 S2. Expert Elicitation Instrument S3 S3. Additional Comments on Expert Elicitation Instrument Design S3 S4. Brief Analysis of Cognitive and Motivational Biases S5 S5. Insights from the Methodology S6 S6. Results on Overnight Capital Cost for Gen III/III+ and SMRs S8 S7. Returns to RD&D Results S9 S8. Distribution of Nuclear RD&D Budget Recommendations S10 S9. Additional Discussion on RD&D Thrusts and Gen. IV Reactor Systems S10 S10. Results on Nuclear Plant Risk Premium Compared to Natural Gas Plants S11 S11. Results on Constraints on Future Nuclear Energy Growth S12 S12. Results on the Impact of Fukushima on Experts’ Opinions S15

SI References S15

  S2

S1. List of the 60 E.U. and U.S. Nuclear Experts Who Participated in this Study Table S1: Experts who participated in the online survey; in grey, experts who participated in the workshop. Name Previous and/or current affiliation Country Name Previous and/or current affiliation Country

Markku Anttila

VTT (Technical Research Centre of Finland) Finland John F. Ahearne

National Academy of Sciences, Sigma Xi, Nuclear Regulatory Commission

USA

Fosco Bianchi

Italian National Agency for New Technologies, Energy & Sustainable Economic Development (ENEA)

Italy Johnhong Ahn

University of California, Berkeley USA

Luigi Bruzzi University of Bologna Italy Edward D. Arthur

Advanced Reactor Concepts, Los Alamos National Laboratory, University of New Mexico

USA

Franco Casali

ENEA; IAEA; University of Bologna Italy Sydney J. Ball

Oak Ridge National Laboratory USA

Jean-Marc Cavedon

Paul Scherrer Institut Switzerland Ashok S. Bhatnagar

Tennessee Valley Authority Nuclear Operations USA

Didier De Bruyn

SCK CEN, the Belgian Nuclear Research Centre

Belgium Robert J. Budnitz

Lawrence Berkeley National Laboratory, Nuclear Regulatory Commission

USA

Marc Deffrennes

European Commission, DG TREN, Euratom Belgium Douglas M. Chapin

MPR Associates USA

Allan Duncan

Euratom, UK Atomic Energy Authority, HM Inspectorate of Pollution

United Kingdom

Michael L. Corradini

University of Wisconsin-Madison USA

Dominique Finon

Centre National de la Recherche Scientifique (CNRS), Centre International de Recherche sur l’Environnement et le Developpement

France B. John Garrick

U.S. Nuclear Waste Technical Review Board USA

Konstantin Foskolos

Paul Scherrer Institut Switzerland Michael W. Golay

Massachusetts Institute of Technology USA

Michael Fuetterer

Joint Research Centre - European Commission The Netherlands

Eugene S. Grecheck

Dominion Energy USA

Kevin Hesketh

UK National Nuclear Laboratory United Kingdom

Pavel Hejzlar

TerraPower USA

Christian Kirchsteiger

European Commission, DG Energy & Transport

The Netherlands

J. Stephen Herring

Idaho National Laboratory USA

Peter Liska Nuclear Power Plants Research Institute Slovak Republic

Thomas Isaacs

Stanford University, Lawrence Livermore National Laboratory

USA

Bruno Merk Institute of Safety Research Forschungszentrum Dresden-Rossendorf

Germany Kazuyoshi Kataoka

Toshiba USA

Julio Martins Montalvão e Silva

Instituto Tecnologico e Nuclear Portugal Andrew C. Klein

Oregon State University USA

Stefano Monti

Italian National agency for new technologies, Energy and sustainable economic development (ENEA)

Italy Milton Levenson

Oak Ridge National Laboratory, Bechtel, EPRI USA

William Nuttall

University of Cambridge

United Kingdom

Regis Matzie

Westinghouse USA

Francois Perchet

World Nuclear University United Kingdom

Andrew Orrell

Sandia National Laboratory USA

Enn Realo Radiation Safety Department, Environmental Board, Estonia; University of Tartu

Estonia Kenneth L. Peddicord

Texas A&M University USA

Hans-Holger Rogner

International Atomic Energy Agency (IAEA) Austria Per F. Peterson

University of California, Berkeley USA

David Shropshire

Joint Research Centre - European Commission The Netherlands

Paul S. Pickard

Sandia National Laboratory USA

Simos Simopoulos

National Technical University of Athens; Greek Atomic Energy Commission, NTUA

Greece Burton Richter

Stanford University, Lawrence Livermore National Laboratory

USA

Renzo Tavoni Italian National agency for new technologies, Energy and sustainable economic development (ENEA)

Italy Geoffrey Rothwell

Stanford University USA

Andrej Trkov Institute Jozef Stefan Slovenja Pradip Saha

GE Hitachi Nuclear Energy USA

Harri Tuomisto

Fortum Nuclear Services Oy Finland Craig F. Smith

Lawrence Livermore National Laboratory, Monterey Naval Postgraduate School

USA

Ioan Ursu Horia Hulubei National Institute of Physics and Nuclear Engineering (IFIN-HH)

Romania Finis Southworth

Areva North America USA

Bob van der Zwann

Energy Research Centre of the Netherlands (ECN)

The Netherlands

Temitope Taiwo

Argonne National Laboratory USA

Georges van Goethem

European Commission, DG Research, Euratom

Belgium Neil E. Todreas

Massachusetts Institute of Technology USA

Simon Webster

European Commission, DG Energy, Euratom Belgium Edward Wallace

NuScale, PBMR Ltd. , Tennessee Valley Authority

USA

  S3

S2. Expert Elicitation Instrument Link to U.S. nuclear elicitation survey: https://erd3energystudy.org/KSG_forms/index.php Username: ES&T Password: TtfHyx Link to E.U. nuclear elicitation survey: https://erd3energystudy.org/icarus/index.php Username: ES&T Password: tgk5Fe S3. Additional Comments on Expert Elicitation Instrument Design To reduce bias in the experts’ estimates, the introductory section of the elicitation was designed to train the experts on probability and bias concepts. Specific response modes—in particular the use of percentiles to capture uncertainty versus the probability of meeting certain cost goals, the graphic supports for the relationship between cost and RD&D levels, as well as uncertainties in estimates and experts’ budget allocations—were chosen to facilitate a speedy completion and correct interpretation of the elicitation. The feedback from experts on these visual tools was positive. After experts completed the online elicitation, experts received a summary of the answers of all participating experts, giving them a chance to provide feedback. The left panel in Figure S1 summarizes the structure of the process followed to conduct the surveys. The right panel shows a structure of the online elicitation instrument. Experts’ judgments can be influenced by clarity and survey design, as well as by motivational biases caused by incentives or pressures that lead people to provide answers that do not entirely reflect their beliefs. Sources of motivational biases can include, for example, social pressure, from the interviewer or from the group of experts (“group think”). The on-line questionnaires allowed us to avoid the group-think bias. The group workshop allowed us to verify the online survey findings, discussing areas where experts may have had different interpretations of the questions, and allowing the experts to modify their views through discussion and debate. The elicitation instrument included several graphic displays to help experts visualize their answers. For example, experts were able to see their 10th, 50th, and 90th percentile estimates of cost for different RD&D levels in a single graph and had 100 chips (representing percentages of their recommended RD&D investment) that they could allocate in a “board game” that included different nuclear technology areas and innovation stages. The design of the survey involved several choices, including expert selection, the mode of the survey, the structure of the elicitation, the visualization of the answers, and the phrasing of the questions, among others. These choices were made over a couple of months with the help of two nuclear experts, who tested the surveys and provided feedback on everything from terminology to the appropriateness and clarity of questions and the tool.

  S4

(a)

(b) Figure S1: (a) Structure protocol employed in the design of the online elicitation and group discussion; (b) structure of the individual online elicitation instrument. All question formats have their advantages and disadvantages, and there is no agreement regarding the best way of phrasing these questions. In our case, it was necessary to keep the survey to a manageable length (it already took experts over 2 hours to complete it). Within these limitations, we chose a Likert scale to frame the question about the non-cost benefits of RD&D; the literature suggests that this approach is preferable to “forced choice” responses (i.e., true/false) in most cases (38). In addition, we had short responses and an odd number (5) of response choices, since the literature indicates that Likert scales should be odd-numbered whenever they represent a categorical variable, and should, all other things being equal, be shorter rather than longer (38). We also defined the five points in the scale quantitatively to ensure consistency in interpretation among experts. Below are the definitions of the five points in the scale that were provided to each expert in the survey:

  S5

1 = inability to address goal would not significantly constrain the deployment of

additional nuclear facilities in 2030-2050 (less than 10% lower). 2 = inability to address goal would slightly constrain the deployment of nuclear

facilities (between 10-40% lower). 3 = inability to address goal would moderately constrain the deployment of nuclear

power capacity from 2030-2050 (between 40-60% lower). 4 = inability to address goal would significantly constrain the deployment of nuclear

power capacity from 2030-2050 (between 60-90% lower). 5 = inability to address goal would greatly constrain the deployment of additional

nuclear facilities in 2030-2050 (at least 90% lower).

This definition of the points on the scale allowed us to capture the experts’ judgments of RD&D’s potential benefit in addressing constraints on the future growth of nuclear energy. If, however, some experts thought that improvements in a particular factor such as safety would not have great impacts on the future growth of nuclear energy but would reduce the risks of that growth, this benefit might not show up clearly in this scale.

S4. Brief Analysis of Cognitive and Motivational Biases The online survey was designed to minimize the occurrence of cognitive biases in the experts’ probability estimates. Biased estimates might result, for example, from the heuristic procedure of “availability” (linked to the ease with which experts could imagine an event occurring), from the “anchoring effect” (by which the experts might choose an initial anchor and then adjust their estimates around it), or from the “representativeness” of an event (which can sometimes influence its judged probability) (SI Ref. 1). Experts are also often overconfident in the precision of their estimates, underestimating their uncertainty. To address these biases, our survey instrument (available in its entirety online using the information provided in S2 in this SI) included, among other things: (a) a section encouraging experts to think about the lowest possible and highest possible costs and about the range of events that may affect costs before asking them to introduce their 50th, 10th, and 90th percentiles; (b) a background information section to ensure that experts had a range of information available to them to help them overcome anchoring (though the background information could have also provided a common set of anchors, most experts projected costs that were higher than those presented in the background section, indicating that the experts did not anchor strongly on the numbers presented in the background information); and (c) a section explaining overconfidence with a historical example of overconfidence in estimates. Our survey instrument did not directly address availability and representativeness biases in matters such as estimating the risk of an accident or of a terrorist attack, but it is notable that the Fukushima accident, which surely dramatically increased the mental availability of major accidental radiation releases, had only a modest effect on most experts’ estimates of the nuclear future or the future risk of further accidents. This may suggest that availability bias was not a major driver of their initial estimates. We comment on Fukushima’s impact on the experts’ estimates in section S12 of this SI. Among other sources of bias, the experts’ background and expertise is obviously a key driver of their judgment. Motivational biases could have come into play if experts thought that recommending large amounts of RD&D funding for their areas may result in more funds for research in their area. To determine whether experts consistently recommended larger RD&D

  S6

investments for their areas of expertise, we included a section in the survey asking experts to rate their level of expertise on different reactor technologies and cross-cutting research areas. For E.U. experts there is a low and negative Pearson’s correlation (-0.41) between high expertise in an area and high budget allocation. For U.S. experts, the correlation is positive but low (0.2). In short, it does not appear that the experts allocated substantially more RD&D spending to their own areas of expertise. S5. Insights from the Methodology The two-step procedure, soliciting experts individually and then following up in a group discussion, made it possible to identify key issues that could arise when each of the two methodologies is followed as a stand-alone procedure and was a relatively cost-efficient protocol. There are aspects of the individual remote surveys that the workshop helped clarify: - Verification of the online expert elicitation results on costs and performance. Experts

participating in the workshop were asked about the clarity of the questions (particularly their answers about the probability of cost and performance for different RD&D scenarios) and given the opportunity to change their answers. Few U.S. and E.U. experts made changes to their cost answers, and these changes were minimal: one U.S. expert revised his Gen. III/III+ BAU cost estimates slightly upward, another U.S. expert revised his Gen. IV cost estimates (also slightly upward), and another U.S. expert revised his BAU SMR cost estimate slightly downward. Two E.U. experts revised their Gen. III/III+ costs slightly (one up and one down), two E.U. experts revised their Gen. IV costs slightly (one up and one down), and one E.U. expert revised his SMR cost estimate downward. These changes were minor and were a response to new information that became available between the time when experts filled out the individual survey and the workshop, which occurred in the backdrop of the Fukushima nuclear accident in March 11, 2011. The discussion at the workshop itself did not appear to lead to significant changes in experts’ projections.

- Impact of Fukushima on the future of nuclear deployment in the United States and the European Union. In contrast with the cost estimates, which remained largely unchanged after the workshop, E.U. and U.S. experts adjusted the probabilities that they had assigned to three scenarios for the future of nuclear power in their region, as discussed below.

- Impact of workshop discussion on individual judgments on nuclear RD&D thrusts. The discussion at the workshop appears to have contributed to increased consensus among the experts on RD&D allocation. RD&D areas that gained in importance after the U.S. and E.U. experts revised their answers were SMRs, risk and safety, and proliferation resistance. E.U. experts also increased their recommended allocations for SFRs and fuels and materials, while reducing the allocations for MSRs and fuel cycle technologies. U.S. experts allocated slightly lower shares of RD&D investments during the workshop to SCWRs, VHTRs, and SFRs.

- Improved understanding of how some experts perceived questions about events. While experts displayed a clear understanding of the questions asked about costs and performance, during the workshop it became clear that different experts were using a different definition of

  S7

“major radioactivity releases caused by an accident or sabotage” when asked about the impact that such an event would have on the growth of nuclear power, and the probability of such an event occurring. While some thought that the Fukushima accident would fall under their definition of a “major radioactivity release,” others felt that such a description would only apply to a larger accident with more direct casualties.

- Improved understanding of motivations behind experts’ RD&D recommendations. The discussion at the workshop revealed that while some experts thought of climate change mitigation as the main goal when making RD&D recommendations, others had multiple goals in mind. This variation in the experts’ reasoning would not have been revealed had we pursued only the individual elicitations.

- Clarity on reasons behind differences between the emphasis of E.U. and U.S. experts. U.S.

experts place more emphasis on RD&D to clarify fuel cycle economics and reduce fuel cycle costs than E.U. experts do. During the workshop experts explained that this is likely to be caused by the fact that in the United States there is a greater focus on private sector involvement. Large-scale fuel cycle facilities would likely not be built in the United States unless private firms believed it was profitable to do so, while in Europe such activities might be undertaken by state-owned firms or with other state support.

- New insights on differences of opinion between U.S. and E.U. experts. Even though the ranges of estimates of the future costs of SMRs were similar between experts across the Atlantic, E.U. experts thought it unlikely that a substantial market for SMRs would develop, while U.S. experts were split between SMR market optimists and skeptics. There are several reasons for this, including the fact that financing GW-scale nuclear power plants in the United States may be more difficult than in some European countries which may have government involvement in nuclear financing.

As stated earlier, while insights on costs did not change during the workshop from the individual expert elicitations, the workshop enriched the information obtained from the elicitations on other topics.

  S8

S6. Results on Overnight Capital Cost for Gen III/III+ and SMRs Figures S2 and S3 present the expert estimates for Gen. III/III+ and SMR reactors respectively.

Figure S2: Gen. III/III+ overnight capital cost estimates of U.S. (top panel) and E.U. (bottom panel) experts in 2010 (gray squares), 2030 under BAU RD&D (black squares), 2030 under recommended RD&D (black square outlines), and 10th and 90th percentile error bands (vertical lines). The range of AFC Cost Basis estimates is also provided for comparison (horizontal lines in gray) (SI Ref. 2). Data also presented in SI Ref. 3.

Figure S3: SMR 2030 overnight capital cost estimates of U.S. (top panel) and E.U. (bottom panel) experts under BAU RD&D (black squares) and under recommended RD&D (black square outlines). 10th and 90th percentile error bands (vertical lines). Data also presented in SI Ref. 3.

  S9

S7. Returns to RD&D Results Figure S4 shows that experts expect decreases in overnight capital cost of the three systems investigated resulting from RD&D alone, excluding learning-by-doing in building subsequent plants.

Figure S4: Impact of RD&D investments on the overnight capital cost reduction in 2030 of Gen. III/III+ plants (triangles, top panel), Gen. IV plants (diamonds, middle panel), and small modular reactors (squares, bottom panel). Note that the cost reductions are associated with RD&D investment in either the United States or the European Union as a whole. Table S2 shows the results of the returns to RD&D regressions.

Table S2: Estimated impact of public RD&D to reduce overnight capital costs in 2030 when compared to costs in 2030 under a BAU public RD&D funding level.

Parameter Gen. III/III+ Gen. IV SMR

Returns to R&D rate 1.36% 5.13% 5.11%

p-value 0.001 0.000 0.000

R2 0.1065 0.2644 0.2714

  S10

S8. Distribution of Nuclear RD&D Budget Recommendations Figure S5 shows the recommended investments in annual public nuclear RD&D investments for the European Union and the U.S. federal government from E.U. and U.S. experts, respectively.

Figure S5: Distribution of experts’ recommended total RD&D budget to nuclear energy.  

S9. Additional Discussion on RD&D Thrusts and Gen. IV Reactor Systems Experts were also asked to write down the RD&D objectives they intended to achieve in the four areas to which they devoted the largest amount of funds. Here we provide some of the key objectives and differences between U.S. and E.U. experts. Experts devoting significant funds to SMRs described several benefits to these investments, including: less lumpy capital investments, the possibility to achieve economies of scale in manufacturing (to possibly outweigh the loss of economies of scale in power generation), increased siting flexibility, a reduction in construction times, and the possibility of recycling existing sites. US experts emphasized the need for support for demonstration and safety testing, while E.U. experts emphasized the need for longer-term R&D on fuels and materials that would enable particular types of SMRs. U.S. experts recommending large amounts of funds for fuel cycle research emphasized research to improve economics, while E.U. experts emphasized waste minimization. Both U.S. and E.U. experts mentioned separation chemistry of minor actinides and research on spent fuel geologic disposal and high-level waste repositories as important RD&D objectives. Both U.S. and E.U. experts emphasized research on materials for reactor cores, valves, pumps, intermediate heat exchangers, etc. (e.g., ceramic composites and nanomaterials); fuels to withstand extreme temperatures and high-radiation fields; and research on improved simulation codes. Each Gen. IV reactor type is said to have different advantages. In essence, the fast reactors have the ability to breed more plutonium (or U-233) than they consume, so if limited uranium resources are a major concern, they could help address that problem. But it will probably be many decades, and perhaps longer, before uranium supplies become limited enough for this to be

  S11

a major driver of reactor deployment choices. Fast reactors also have the potential for improved transmutation of long-lived radionuclides, possibly easing the waste management problem. The sodium-cooled fast reactor is the most popular of the fast reactors because there is the most experience with it, and sodium has some desirable properties as a coolant. The lead-cooled or gas-cooled fast reactors could get the uranium-resource extension of fast reactors without sodium’s fire risks — but with new problems for their coolants. The high-temperature gas reactor offers the possibility of increased inherent safety (including the ability to survive a cutoff of all cooling without melting down). That system also goes to very high temperatures, which increases efficiency and offers the opportunity to provide high-temperature process heat, which could improve overall economics. The molten salt reactor has no solid fuel (so no fuel fabrication costs and no danger of fuel melting), and offers the opportunity to separate out a portion of the fission products on each cycle, breeding uranium or plutonium without ever separating out weapons-usable nuclear material from the rest of the molten salt. As shown in Figure S6, U.S. experts are generally more optimistic regarding the commercialization date of Gen. IV systems.

 Figure S6: Time when a Gen. IV nuclear plant or system will first become commercially available under the RD&D scenario proposed by expert.  S10. Results on Nuclear Plant Risk Premium Compared to Natural Gas Plants Experts were asked to evaluate the risk premium for nuclear power facilities over that of natural gas power plants. Risk premium is the financial return in excess of the risk-free rate that a riskier investment is expected to yield, to attract capital from investors. As shown in Figure S7, most experts thought that, in the short term (around 2010 or the present time), nuclear projects would have to be pay a substantial risk premium compared to natural gas-fired power projects, to account for factors such as possible delays, cost over-runs, or project cancellations. As time goes on, however, most experts expect this risk premium to decline (though not to zero). Some of the estimates of the risk premium by E.U. experts appear unrealistically high (at 20-50%, they are higher than most estimates of the total capital charge for financing nuclear plants); these experts may have interpreted the question as referring to the fractional increase in the capital charge, so that if the financial return required for nuclear energy was 12% rather than 10% for natural gas (a 20% higher figure), the risk premium would be 20%.

0

2

4

6

8

10

12

14

16

2020 2025 2030 2035 2040 2045 2050 2055 2060 2065 2070 2075 2080 2085 2090

Num

ber

of

exp

ert

s

Year of commercialization

U.S. expert

E.U. expert

  S12

Given that financing costs are a key factor undermining the competitiveness of nuclear energy, it is striking that these nuclear technology experts, almost all of whom are nuclear energy advocates, overwhelmingly predicted that as far in the future as 2050, nuclear energy would still be considered a somewhat riskier investment than other types of electric power.

Figure S7: Risk premium for nuclear investment above natural gas power plant discount rate according to E.U. experts (above) and U.S. experts (below). Data also presented in SI Ref. 3. S11. Results on Constraints on Future Nuclear Energy Growth A wide range of factors could affect the future of nuclear energy, only some of which are susceptible to being addressed by RD&D investments. We therefore asked experts about what factors and what future events would be most important in promoting or constraining nuclear energy growth, and about the probability of different future growth scenarios in their regions. (Our survey did not ask the experts to judge the likelihood of different scenarios for global nuclear energy deployment; this would be a useful area for further research.) Overwhelmingly, the negative future event that the participating experts expected would have the largest impact on nuclear energy growth was an accident or terrorist attack somewhere in the world leading to a major radioactive release. Both U.S. and E.U. experts generally estimated that such an event would reduce future deployment in their region by 50-100%. Most experts, however, judged the probability of such an event occurring anywhere in the world to be 10% or less over the next 20 years. (Only one expert participating in the workshop changed his estimate on this point after Fukushima.) The other events the participating experts believed would have the largest negative impact were major cost overruns in nuclear projects or a cascade of proliferation perceived as resulting from the civilian nuclear energy system. Many experts believed that major cost overruns were fairly likely (with roughly half of the experts putting the likelihood of such overruns above 50%). Both U.S. and E.U. experts thought that successful siting and operation of nuclear waste

  S13

repositories in the United States and in other countries would be a major step forward for nuclear energy, increasing likely growth significantly, and both thought such successes were fairly likely, with other countries being even more likely to succeed than the United States (see Figure S8).

Figure S8: Figure S8a shows the percentage change in future nuclear growth in their region U.S. and E.U. experts project would result from various future events. Figure S8b shows the experts’ judgment as to the probability of those events. The numbers in each box are the number of experts providing that rating; ratings with the largest concentrations of experts are shaded. Data also presented in SI Ref. 3.

  S14

Figure S9: Experts’ probabilities for different diffusion scenarios in 2050 (top panel for E.U. experts, bottom panel for U.S. experts). For the U.S. experts, the three scenarios evaluated are: Low: maintaining current U.S. nuclear power capacity (100.6 GWe) in 2050; Medium: 286 GWe of U.S. nuclear capacity in 2050 (which corresponds to the low scenario of MIT’s 2003 Future of Nuclear Power report (SI Ref. 4)); 477 GWe of U.S. nuclear capacity in 2050 (the high scenario of MIT’s 2003 Future of Nuclear Power report). For the E.U. experts, the three scenarios evaluated are: Low: maintaining current E.U. nuclear power capacity (170 GWe) in 2050; Medium: 280 GWe of E.U. nuclear capacity in 2050 (half the growth rate of the IAEA 2006 Energy, Electricity and Nuclear Power Estimates (SI Ref. 5)); High: 400 GWe of E.U. nuclear capacity in 2050 (the growth rate of the IAEA 2006 Energy, Electricity and Nuclear Power Estimates). On average, the participating experts were cautiously optimistic about future nuclear energy growth in their regions, though there were sharp differences of view. When presented with a low (no growth), medium, and high scenario for nuclear growth in their regions, only a few thought the probability of the low scenario was 70% or more, but there were also only a few who thought it could be ruled out. The same held for the high-growth scenario. The medium scenario (defined as 280-290 GWe of nuclear capacity installed by 2050 both in the United States and in the European Union) had the highest probability estimate, averaging across all the experts. Overall, the vast majority of both U.S. and E.U. experts projected that it was more likely than not

  S15

that nuclear energy would achieve either medium or high growth in their regions by 2050 (see Figure S9). S12. Results on the Impact of Fukushima on Experts’ Opinions  

Participating experts provided their on-line responses well before the Fukushima accident, while the workshop took place four weeks after the accident began. Experts participating in the workshop had an opportunity to discuss their answers and the issues arising from the Fukushima accident in-depth, and were given an opportunity to modify their previous judgments if they so chose. Only a few experts chose to modify their projections of nuclear power costs or their recommendations with respect to nuclear RD&D investments. Both E.U. and U.S. experts, however, adjusted their projections for nuclear power growth after Fukushima. Three out of the six U.S. experts participating in the workshop reduced their estimates of the probability of high nuclear growth – from 60% to 30% in two cases and from 35% to 20% in another case. One of the remaining three U.S. experts adjusted his probability for the high growth scenario slightly upward, from 35% to 40%. Four of the twelve E.U. experts participating in the workshop cut their estimates of the probability of high growth – from 45%, 60%, 20%, and 30% to 20%, 50%, 10%, and 10%. The experts participating in the workshop generally agreed that (a) as of late April 2011, it was too soon to draw many of the lessons from Fukushima until more was understood; (b) there would inevitably be tightened regulation in several areas and increased emphasis on safety; and (c) with appropriate government investment and support, nuclear power could still achieve large-scale global growth by 2050. The participating experts argued that while a few existing nuclear power countries (such as Germany) might change course, and some “newcomer” countries considering their first plant might reconsider, the largest nuclear markets (such as China) were likely to continue on a path of substantial growth. The discussion at the workshop appears to have contributed to increased consensus among the experts on RD&D allocation. RD&D areas that gained in importance after the U.S. and E.U. experts revised their answers were SMRs, risk and safety, and proliferation resistance. E.U. experts also increased their recommended allocations for SFRs and fuels and materials, while reducing the allocations for MSRs and fuel cycle technologies. U.S. experts allocated slightly lower shares of RD&D investments during the workshop to SCWRs, VHTRs, and SFRs. SI References (SI Ref. 1) Tversky, A.; Kahneman, D. Judgment under uncertainty: Heuristics and biases. Science, 1974, 185:1124-1131. (SI Ref. 2) Shropshire, D. E.; Williams, K. A.; Hoffman, E. A.; Smith, J. D.; Hebditch, D. J.; Jacobson, J. J.; Morton, J. D.; Phillips, A. M.; Taylor, J. P. Advanced Fuel Cycle economic analysis of symbiotic light-water reactor and fast burner reactor systems. Technical Report from the Idaho National Laboratory. Document INL/EXT-09-15254, U.S. Department of Energy, Washington D.C., United States. January 2009. (SI Ref. 3) Anadon, L.D.; Bosetti, V.; Bunn, M.; Catenacci, M.; Lee, A. International Workshop on Research, Development, and Demonstration to Enhance the Role of Nuclear Energy in Meeting Climate and Energy Challenges. Report for Energy Technology Innovation Policy research group, Belfer Center for Science and International Affairs, Harvard Kennedy School,

  S16

ICARUS Project, Fondazione Eni Enrico Mattei, and International Center for Climate Governance, April 2011. (SI Ref. 4) MIT. The Future of Nuclear Power- An Interdisciplinary MIT Study. Massachusetts Institute of Technology. Cambridge, Massachusetts, United States. 2003. (SI Ref. 5) IAEA. Energy, Electricity, and Nuclear Power Estimates for the period up to 2030. (International Atomic Energy Agency. Vienna, Austria). July 2006.