Cost savings from nuclear regulatory reform: an econometric model.

I. Introduction

The nuclear-generated power touted in the 1950s as someday being "too cheap to meter" got dismissed in the 1980s as incapable of being both safe and cost effective. Today, less than 20 percent of American's electricity is nuclear-generated, no new plants are planned or on order, and some of the earliest units are scheduled for decommissioning within the next decade. Even so, interest in nuclear power has been revived by increasing energy demands, concerns about global warming, and the uncertainty surrounding oil resources in the Persian Gulf. As a long-term alternative to fossil fuels, atomic energy offers the important advantages of clean air and domestic availability of fuel. But these advantages will count for little unless and until the costs of nuclear power can be seen as reasonable.

Our premise is that the relevant costs are those of providing safe and environmentally clean electric energy. To the extent that increased costs have resulted from increasingly stringent regulations, they reflect the internalization of external costs. Indeed, the external costs of nuclear power (particularly safety and environmental protection) have been internalized to a greater degree than with most alternative fuel sources used by electric utilities. Nuclear construction costs are properly compared with those of alternative sources only after the latter are adjusted for environmental damage and endangerment, including, as examples, the costs of oil spills, of building double-hulled tankers, and of building off-shore offloading facilities. A shift to nuclear sources could reduce these costs whereas it would increase disposal costs for radioactive materials. We do not estimate any of these environmental costs in this article; each would be a major study in itself. We nonetheless contend that a better understanding of nuclear plant construction costs is pivotal to a balanced evaluation of the merits of uranium relative to other fuel choices.

Several regulatory reforms have been under consideration. At the federal level, the utilities are currently seeking to streamline and expedite the licensing process in an effort to abbreviate construction durations. Utilities are proposing pre-approved, standardized designs, pre-approved sites, one-stop licensing and legislative-style NRC hearings instead of adjudicatory proceedings. At the state level they propose the elimination of retroactive prudence reviews [11]. Our study provides insights into the effects of these proposals.

Despite the public clamor over nuclear power costs, the population of related articles in the economic literature is small. In a study by Mooz [5], nuclear construction costs are related to significant regional effects, scale economies and learning effects. Mooz also finds higher costs related to cooling towers in the plant's design. A follow-up study [6], with an expanded data base, confirmed learning effects but failed to find scale economies. Zimmerman [12] examines learning-by-doing effects both for the firm and the industry, with internal learning reducing costs at nearly twice the rate of external learning. He finds that construction costs rise in a one-to-one proportion to duration (construction start date to commercial operation date). Komanoff [3] and Cantor and Hewlett [1] respectively find reductions in overnight costs of 13 percent and 47 percent upon doubling plant size. Both studies nonetheless find scale economies are largely offset by a rise in duration associated with project size.

A U.S. Department of Energy (DOE) study [2] distinguishes resources costs and "time related" costs, attributing 75 percent of construction cost to increases in the former. Duration, however, adds to resources cost because it increases with (unmeasured) design changes as well as with safety and environmental retrofits required by regulations. DOE finds significantly higher costs for utilities using outside firms for design and engineering than for utilities performing these functions themselves. Thus, internal experience has a positive learning effect. Radlauer, Bauman, and Chapel [9] attribute a shift in construction costs from the pre-1972 era to the post 1971 era to an abrupt change in the regulatory environment.

Our model has several unique features. (1) The cost data are carefully adjusted to limit the number of remaining explanatory variables, increasing the degrees of freedom and minimizing multicollinearity. (2) Economies of scale for nuclear projects are correctly identified for the first time. (3) Learning effects are detected by an experience variable. (4) A duration variable is introduced to measure the effects of delay. Our findings comprise a bridge across the earlier findings [1; 2; 5; 6; 8; 12].

II. Theory

The real cost of nuclear power projects has greatly increased since the first plant became operational in 1957. Theory suggests three important contributors to this trend: increased regulatory requirements, changing market conditions, and lengthening construction durations.

Designed to ensure the reliability and safety of output, regulations since at least the 1960s have tended to be increasingly stringent. Their impact will differ among plants, depending on the efficiency with which each utility and architect/engineer deal with them. We nonetheless can expect that plants completed under later regulatory requirements will tend to cost more than those built earlier.(1) The consequent engineering features (e.g., a cooling tower requirement) may account for a sizable part of the cost differences among nuclear power plant construction projects. However, the cost effects of NRC regulation should not be allowed to mask the effects of mismanagement or incompetence. If poor design or construction work require portions of the plant to be torn out and rebuilt, the resulting delays and increased costs may owe more to poor management than to the NRC. Therefore, project duration is an important cost variable. Unusually lengthy construction indeed may indicate a comparatively inefficient, ineffective, or incompetent project team. A short duration can imply that the team managing and building the plant is particularly effective. Presently, each U.S. plant is unique; however, intuition suggests that the experience variable would have higher values and construction duration would be significantly reduced if the industry were to repeatedly build one or two standard designs.

Demographics also play a role in nuclear plant costs. A small cost differential is attributable to climate, since very cold or wet weather can impede construction. Urban congestion can also add to plant costs, creating difficulties in the shipping of materials or forcing the imposition of additional safety requirements. Because it is often difficult to isolate the various contributors to these geographically related costs, a categorical variable is sometimes used, capturing the significance of all regional factors combined. A more sophisticated approach - the one we use - is an accounting of known wage and materials cost differentials. When adjusting the plant cost data for inflation (putting all plants into equivalent real terms), we also adjust for regional differences in cost of living and other costs of construction.

Learning based upon project design experience can lower costs. As a utility gains experience in the hiring and use of architect and engineering firms, it may gain a learning-by-doing effect in design efficiency. Construction efficiency is closely related both to design efficiency and to technological improvements shared by the entire power plant construction industry.

Economies of scale can also lower plant costs. Beyond some point, however, the scale economy diminishes as larger volumes come to require extra structural reinforcement, special materials, unusual construction techniques, or complex management or monitoring systems.

Finally, whereas there is a cost to regulatory standards, there also is a benefit. Output in terms of Mwe capacity is qualitatively improved by design enhancements (rising regulations). We assume, in the interests of simplicity, that regulations effect output in such a way that relative input shares remain constant along the expansion path (Hicksian neutrality). Paik and Shriver [8] find some evidence supporting Hicksian neutrality. The most general construction production function would be

q = G(R)F(L, K), (1)

in which labor and capital are adjusted by a function of the regulatory requirements or G(R). By the same token, the value of q embodies an environmentally cleaner and safer capacity. If other inputs such as materials are affected by regulation, we have

q = G(R)H(L, K, M), (2)

where M is a vector representing other inputs.

According to duality theory, we can be indifferent between estimation of the production or of the cost relationship. Either would provide full explanation of the structure of production. We select the cost function because it provides a more reliable estimate of the effects of exogenous forces and because regulation, learning-by-doing, and managerial inefficiency in the architectural engineering stages are directly related to construction inputs rather than to construction output.

The most general dual cost function (as related to equation(2)) is

C = J(q, N, R, P, Z), (3)

where q is net reliable, safe electric generating capacity (Mwe), N is an index of learning-by-doing, R is a measure of regulation, P is a vector of input prices, and Z is a vector of special design choices.

Sometimes it is useful to "deflate" or "inflate" the dependent variable by known prices of inputs or other price differentials. Since nuclear construction projects are of long duration and are completed over different periods and with different variable price sets, an alternative version of equation (3) would be

L[(X).sup.-1]K[(P).sup.-1]C = J(q, N, R, Y), (4)

in which K is a function of the vector of input prices over time, L is a function of X which is a subset of Z, and Y represents the residual subset of Z. Then equation (4) can be simplified to

[C.sup.*] = J(q, N, R, Y), (5)

where [C.sup.*] is the adjusted cost variable.

From these equations, we estimate the parameters for two models. The initial choice for the cost equation is a multiplicative form in order to capture the interaction of variables and the expectation that cost increases due to duration and decreases due to size are nonlinear. Therefore, in Model A, ordinary least squares (OLS) is applied to

In [C.sup.*] = [B.sub.0] + [B.sub.1] lnR + [B.sub.2] ln q + [B.sub.3]DUR + [B.sub.4]N, (6)

where [C.sup.*] is adjusted construction cost per kilowatt hour (details below), R is the number of total regulations up to the completion date of the last unit of a project, DUR is the number of months between construction start and commercial operation, and N is the number of previous nuclear projects owner-operated by the utility prior to the construction start date (the experience variable).

In Model B, OLS is applied to

In [C.sup.*] = [C.sub.0] + [C.sub.1] lnR + [C.sub.2] lnR([D.sub.1]) + [C.sub.3] lnq + [C.sub.4] lnq([D.sub.2]) + [C.sub.5]DUR + [C.sub.6]DUR([D.sub.3]) + [C.sub.7]N + [C.sub.8]N([D.sub.4]). (7)

This estimation allowed us to use the entire sample but distinguish values of the coefficients for two separate periods. The dummy variables (the Ds) were used to determine whether the slope coefficients differed between the two eras selected.

III. Data Issues in Nuclear Plant Comparisons

Our selection of a well-defined population avoids potentially thorny statistical problems. We include in the study all plants whose construction began after 1966 and was completed by 1987. Thus, the small number of projects built in the pre-1967 turnkey period are excluded; they were somewhat experimental in nature, and their costs were partially subsidized by manufacturers who were anxious to promote the new technology. These criteria provide a sample comprised of 53 projects.

By considering the total costs of each construction project, we avoided problems with inconsistent accounting treatments of the common costs of multiple units. By adjusting the cost data to reflect the amount of disallowances for imprudence as established by regulators, we ensured that the model would provide better estimates of the "reasonable" costs of all projects in the population.(2) We updated the Tennessee Valley Authority(3) data with more recent cost and schedule estimates obtained directly from the utilities.

Initially, the construction costs in each year are reported in current dollars of those years. Over time, of course, these are not the same dollars. The industry has developed the idea of "overnight" dollars as a method for expressing costs in terms of comparable dollars. Thus, the nominal outlays for each project, once estimated, must be inflated over time to express costs in today's like dollars ("overnight" dollars). This is done with the regional Handy-Whitman nuclear power plant construction cost index that measures cost of labor, material, and equipment.(4) The overnight costs ([C.sup.*]) are

[C.sup.*] = [summation of]([C.sub.[Tau]]/[I.sub.[Tau]]) where [Tau] = s to f, (8)

where [I.sub.[Tau]] is the Handy-Whitman nuclear power plant construction index, s = start date of construction and f = finish date of construction. The index of construction ([I.sub.[Tau]]) provides a sequence of annual inflation rates where

[Mathematical Expression Omitted],

for t = s + 1, s + 2, . . ., f. The concept of overnight costs is a variation of present discounted value, differing in appearance only because time-f dollars comprise the numeraire rather than time-s dollars.

A meaningful comparison of project costs requires distribution of cash expenditures over the construction cycle for each period. To accomplish this task we applied an industry-standard S curve, a formula used by the Energy Information Administration [2, 89]. The pattern is established by

[y.sub.t] = [{1 - [[cos([Pi]/2)([x.sub.t])].sup.4.08198]}.sup.3.24948] (10)

where [y.sub.t] = cumulative percent of total cost in year t and [x.sub.t] = fraction of total elapsed construction time. The proportion or share spent during year t is [k.sub.t] = [y.sub.t] - [y.sub.t-1]. In turn, these shares for each year are used to derive the annual costs in equation (8). With this normal pattern, total nominal project expenditures can be spread year by year over the entire construction period.(5)

Table I. Nuclear Power Plant Disallowances (000,000)

Plant                  Disallowance    Plant
Disallowance

Byron 1                   $101.5       River Bend         1,400.0
Callaway 1                 440.0       San Onofre 2         344.6
Fermi 2                    397.0       Shearon Harris       322.8
Hope Creek                 489.2       Shoreham           1,395.0
Limerick 1                 368.9       Vogtle 1             951.0
Nine Mile Point 2        1,803.0       Waterford 3          284.0
Palo Verde                  60.0       Wolf Creek           257.3
Perry 1                    628.0

Sources: Commonwealth Edison, Annual Report, 1985; Illinois Power
Company, Kansas City Power & Light Company, Louisiana Power & Light
Company, Philadelphia Electric Service Company, Southern California
Edison Company, Union Electric, Annual Reports, 1986; Gulf States
Utilities, Annual Report, 1987; Michigan Public Service Commission,
Final Order, Case U-7760; New York Public Service Commission,
Opinion No. 86-24, Case 29124; Texas Public Utilities Commission,
Final Order, Docket No. 7460; Moody's Public Utility Manual, 1987;
NARUC Bulletin, January 6, 1986; NASUCA Newsletter, March 1988.

Mooz [5] concluded that Northeastern region plants are the most expensive. Possible causes include a denser population, fewer suitable plant sites, and - most notably - higher construction wage rates. We adjusted the cost of all plants for differences in wage and material costs in each of the six Handy-Whitman regions.(6) By this means costs are pre-adjusted for input price differentials between regions (see equation (4)). This conversion of all plant costs to regionally equivalent overnight dollars permits use of the largest possible sample. The unique conversion ratio for each estimate is dependent primarily on the timing and duration (as well as the regional location) of each construction project.(7)

As the sample includes not only prudent projects, but also projects for which regulators have disallowed certain costs due to mismanagement and inefficiency, we adjusted the cost of imprudent plants by the value of known disallowances, thereby putting the projects on a common basis, as if all were constructed in a prudent manner. This adjustment is not perfect because regulators often lack sufficient evidence to quantify the excessive costs.(8) The plants and disallowances used appear in Table I.

IV. Results

We reduced the set of explanatory variables to the smallest number that will adequately explain construction cost per kw of capacity. Prior to the regression analysis, costs were adjusted for substantial regional differences, disparate inflation rates at different times, and imprudent or unreasonable costs. The reduction in explanatory variables increases degrees of freedom and reduces multicollinearity. In particular, we no longer need region or inflation variables. These adjusted project costs appear in Table II.

We also tested several potential measures of the unique design characteristics of each project (e.g., the presence of cooling towers, the depth to bedrock, the type of reactor vessel). These were not consistently significant, probably because the cost of various design features varies with the method chosen by the architect/engineer. Our final equations thus include no design variables. We identified proxy variables for the remaining explanatory variables.

Descriptive statistics are reported in Table III. The final equation for the first test (Model A) is simple but powerful, containing four independent variables, all significant at the 0.0001 level. The regression results appear in Table IV for both models. We tested for heteroscedasticity, with negative results.(9) The signs and magnitude of all the coefficient estimates accord with our theoretical expectations. Regulatory requirements have a positive effect on construction cost, as the external impacts on safety and the environment have been internalized. Also, in a simulation from the model, cost per kw declines as project size increases, providing the textbook-like graph reflecting economies of scale [ILLUSTRATION FOR FIGURE 1 OMITTED]. The experience variable shows large and favorable learning-curve effects. Construction duration also is significant. The greater the number of months between the construction start date and commercial operation of the final unit, the greater the cost per kw. This result fits our expectations: projects with long durations tend to have very high costs per kw.(10)

Radlauer, Bauman, and Chapel [9] have suggested that the structure of pre-1972 nuclear plant costs is different from that of post-1971 costs, attributing the differences to changes in the regulatory environment commencing with the Calvert Cliffs decision in 1972. While there are undoubtedly differences between the two regulatory environments, a properly specified regulatory variable should account for most increases in cost due to these changes. Hence, we considered it appropriate to use data from both periods.

Nevertheless, we used Model B to test for structural changes between the early (pre-1972) and late (post-1971) eras. In Model B we pooled the two sub-samples and used dummy variables to identify any changes in the slope coefficients between the two eras. In this model, D is a dummy variable having a value equal to unitary for the post-1971 era and zero for the pre-1972 era.(11) The results are reported in Table IV. Most of the nondummy coefficient estimates are significant at the 0.0001 level, a pattern consistent with that of Model A. The slope dummies are generally insignificant, suggestive of no structural differences between the two periods.

[TABULAR DATA FOR TABLE II OMITTED]

[TABULAR DATA FOR TABLE III OMITTED]

The predicted values from Model A are compared with actual values (in 1986 dollars) in Figure 2. As we see, only about five of the actual project costs stray far from their predicted values. Moreover, Model B should not be dismissed entirely. The insignificance of some of the variables results from multicollinearity, according to several standard statistical tests. Model B remains a good predictive model. Each time, when a single slope dummy was introduced into Model A, it was statistically significant at minimally the .1000 level.

V. Policy Simulations

We now return to nuclear policy reform proposals. If a standardized plant size and design were selected, economies of scale advantages close to a "reasonable optimum" could be gained. Then replication of standard designs would cause experience to accumulate quickly. Such homogeneity could speed the construction process greatly. Pre-approved sites, one-stop licensing, and standardized inspection procedures would also serve to reduce duration by simplifying and expediting the regulatory process. To estimate the probable effects of such policy reforms, we used Model A to simulate alternative costs per kw at different values for the independent variables. A policy simulation matrix of the most plausible possibilities within and outside the sample is presented in Table V.

Table IV. Regression Results Dependent Variable: Log of Cost per KW

Independent Variables                    Model A          Model B

Intercept                               5.489150          6.171802
                                       (0.676939)(a)     (1.049015)
                                          0.0001(b)         0.0001

Log of Cumulative Regulations           1.111391          0.847764
                                       (0.097138)        (0.246081)
                                          0.0001            0.0001

Log of Project Net Mwe                 -0.512734         -0.474663
                                       (0.097138)        (0.098971)
                                          0.0001            0.0001

Project Duration                        0.003721          0.005040
                                       (0.001148)        (0.002248)
                                          0.0001            0.0001

Utility Nuclear Project Experience      -0.13427         -0.291892
                                       (0.050057)        (0.136477)
                                          0.0001            0.0150

Regulation Slope Dummy                                    0.085264
                                                         (0.216844)
                                                            0.6502

Size Slope Dummy                                         -0.029657
                                                         (0.137998)
                                                            0.5980

Duration Slope Dummy                                     -0.000109
                                                         (0.002424)
                                                            0.5100

Experience Slope Dummy                                    0.168144
                                                         (0.145461)
                                                            0.8670

n                                             53                53

[R.sup.2]                                  0.893             0.903

F-Ratio                               109.676275         68.652985

a. Standard error.

b. Significance level.

Across the top of the matrix the project size is varied between 800 and 2400 Mwe. The experience variable has been fixed, in turn, at 2 and 3 (within the sample) and then at 4 and 5 (outside the sample values). The value of the experience variable is assumed to increase with standardization because firms would be replicating the same prototype plant. The value of the cumulative regulations is set at its maximum of 153, which assumes no existing rule will be rescinded.

Project duration in our sample ranged from 40 to 213 months, averaging 125. Duration varies (along the vertical) from 60 to 150 in the policy simulation matrix. The larger the size of the project, the lower the costs on each line of the matrix. However, the costs per kw decline by smaller and smaller percentages with each Mwe increase in scale. Moreover, these gains are achieved only if we assume a constant duration. In the real world, of course, duration is not constant. From the sample and from the findings of Komanoff [3] and Cantor and Hewlett [1] we know that duration sometimes increases with the project size. For example, returning to the policy matrix, a 2400-Mwe project completed in 60 months at sample maximal experience (3) costs $1002 per kw (1986 dollars); extended to 150 months it costs $1401 per kw. Often during the 1980s, the larger the project, the greater was the duration and the higher the costs.

Japan's experience suggests that duration need not be closely related to size, according to Navarro [7]. Exceeding the established 600-900 Mwe range raised construction time by only 10 months. Realistically, however, it seems unlikely that an extremely large plant size could be accepted as a standard by the industry, since it would exceed the needs of all but the most densely populated regions. If designs were standardized in the U.S., the optimally-sized project would probably be in the 800-Mwe to 1200-Mwe range. This would ensure usability by both small and large utilities and would allow more frequent repetition of the construction process.

The potential cost savings from more rapid construction is very significant. Duration for the 1979-86 period averaged 61.64 months in Japan versus 140.71 months in the U.S. [7,3-4). Since 1975, the Japanese utility industry has adopted a two-design program, with standardized reactors. In our policy matrix, a 1200-Mwe project would cost $1729 per kw if completed in 75 months by a utility with experience building two previous projects. Increasing the experience value to the sample-maximal value of three reduces the cost to $1512 per kw. If additional experience were gained beyond this historical maximum, the simulated cost would be reduced to $1322 and $1156 as successive projects were built, strongly suggesting the potential benefits of moving down the learning curve through replication. (The outside-of-sample extrapolations can be defended on the grounds of the near-linearity of cost and experience within the sample.)

If a policy of standardization were to achieve these changes in duration and experience, the effect on construction costs would be dramatic - allowing nuclear power to become a viable, low-cost alternative for new baseload electric generating capacity. Pre-approved sites, one-stop licensing, and standardized inspections could reduce costs even more.

Table V. Policy Simulation Matrix: Model A

                                          Project Size (Mwe)
Project Duration (Months)    800     1200     1600     2000
2400

Experience = 2

60                           2013    1635     1411     1258
1146
75                           2128    1729     1492     1330
1212
90                           2250    1828     1577     1407
1281
105                          2380    1933     1668     1488
1355
120                          2516    2044     1764     1573
1433
135                          2661    2161     1865     1663
1515
150                          2813    2285     1972     1759
1602

Experience = 3

60                           1760    1430     1234     1100
1002
75                           1861    1512     1304     1163
1059
90                           1968    1598     1379     1230
1120
105                          2081    1690     1458     1301
1185
120                          2200    1787     1542     1375
1253
135                          2326    1890     1631     1454
1325
150                          2460    1998     1724     1538
1401

Experience = 4

60                           1539    1250     1079      962
876
75                           1627    1322     1140     1017
926
90                           1721    1398     1206     1076
980
105                          1819    1478     1275     1137
1036
120                          1924    1563     1348     1203
1095
135                          2034    1652     1426     1272
1158
150                          2151    1747     1508     1345
1225

Experience = 5

60                           1346    1093      943      841
766
75                           1423    1156      997      889
810
90                           1504    1222     1054      940
857
105                          1591    1292     1115      994
906
120                          1682    1366     1179     1052
958
135                          1779    1445     1247     1112
1013
150                          1881    1528     1318     1176
1071

VI. Conclusions

We use two models in order to estimate the reasonable cost per kw of capacity in the construction of nuclear projects as well as to identify the sources of these costs. The estimated models are statistically reliable and support theoretical expectations. For example, our results confirm that the cost per kw is less if the utility has already moved down the learning curve by building other nuclear plants during this era. Similarly, our results confirm the existence of economies of scale. As expected, the level of NRC regulation has contributed to the cost of nuclear projects.

Duration is critical. Higher costs per kw are associated with less efficient project management and with more complex, more costly designs. Our estimated costs depend heavily on the duration and level of experience assumed so that the policy simulation matrix adds to the case for regulatory reform - a need for pre-approved, standardized designs. Such designs would shorten construction durations and allow the industry to move more rapidly down the learning curve. Since economies of scale increase at a decreasing rate, the "reasonably optimal" size is not necessarily the largest possible; duration can be reduced and experience augmented by the selection of a standard "right-sized" project for regional requirements.

The authors thank an anonymous referee for several helpful suggestions that improved the final product.

1. Some analysts give the impression that NRC regulation is a new phenomenon born of the Three Mile Island accident. This is false; some regulations are as old as the industry. A realistic view of regulation should take into account the overall trend, not dwell on specific incidents.

2. This adjustment of the data does not change the reliability of the parameter estimates. The estimated coefficients and significance levels are similar with or without adjustments for disallowances. This adjustment simply reduces the total costs of construction of projects with disallowances.

3. These data are from the TVA Office of Nuclear Power [10].

4. Individual cost components are weighted to reflect the relative share of each in the construction of a typical nuclear facility.

Marshall and Navarro [4] contend that the use of "overnight costs" contradicts accepted capital theory and such a measure biases the estimate of scale economies upward because "overnight costs" do not include inflation. This argument derives from their misdefinition of "overnight costs": such costs do include inflation. As to accepted capital theory, since rates of return for public utilities in the U.S. are regulated, such regulation encourages utilities to "overbuild" and not to minimize the cost of construction. Thus, accepted capital theory loses much of its relevance when applied to regulated public utilities. Moreover, commissions judging imprudence base the value of disallowances upon "overnight costs." The comparative estimates of what Marshall and Navarro call "true costs" are themselves seriously flawed since they separate first unit and subsequent unit costs, use a "trend" variable as the measure of regulation and fail to include "duration" as an independent variable.

5. Although widely used and accepted, the S pattern will not mirror the precise distribution of expenditures on any specific project; in some cases delays and other problems during the construction may make it a poor approximation indeed. Nevertheless, this drawback must be weighed against the advantages of consistent treatment of all plants in the sample. In any event, the actual monthly patterns of expenditure are not comprehensively available. Hence, an approximation of this sort is not only useful but also necessary.

6. This adjustment does not fully account for the cost savings one would expect in an uncongested area rather than a congested urban area. Nevertheless, it does neutralize many regional cost differences.

7. For consistency, the cost per kw figures for the comparison group excluded AFUDC, since governmental utilities typically do not accrue it (although they do incur financing costs) and there is variability in the AFUDC rates and accrual policies applicable to investor-owned utilities.

8. Regulators subjectively evaluate this phenomena based upon an evidentiary record consisting primarily of thousands of pages of qualitative evidence concerning the history of each individual project. Data limitations force us to draw inferences about the cost of imprudent and inadequate management, rather than measure it directly for each plant in the sample.

9. The generalized least squares generated parameter estimates are not significantly different from those of OLS.

10. The additional cost per kw associated with an additional month of work is particularly important for projects with durations substantially longer than average. These also are the projects that have encountered the most problems. At the extreme, those projects that have encountered massive difficulties tend to have extremely long durations and tend to be the most costly of all, by a wide margin.

11. The complete effect of a particular explanatory variable in this specification is the sum of its regression coefficient and its post-1971 slope dummy coefficient.

References

1. Cantor, Robin A. and James G. Hewlett. "The Economies of Nuclear Power: Further Evidence on Learning Economies of Scale and Cost Estimation." Unpublished manuscript, Oak Ridge National Laboratory, 1987.

2. Energy Information Administration. An Analysis of Nuclear Power Plant Construction Costs. Washington, D.C.: DOE/EIA-0485, 1985.

3. Komanoff, Charles. Power Plant Cost Escalation. New York: Van Nostrand Reinhold, 1981.

4. Marshall, John M. and Peter Navarro, "Costs of Nuclear Power Plant Construction: Theory and New Evidence." RAND Journal of Economics, Spring 1991, 148-54.

5. Mooz, William E. Cost Analysts of Light Water Reactor Power Plants. Washington, D.C.: R-2304-DOE (Rand Corporation), 1978.

6. -----. A Second Cost Analysis of Light Water Reactor Power Plants. Santa Monica, Calif.: R-2794-RC (Rand Corporation), 1979.

7. Navarro, Peter, "Comparative Energy Policy: The Economics of Nuclear Power in Japan and the United States." The Energy Journal, October 1988, 1-15.

8. Paik, Soon and William R. Shriver, "The Effect of Increased Regulation on Capital Costs and Manual Labor Requirements of Nuclear Power Plants." The Engineering Economist, Spring 1981, 223-44.

9. Radlauer, Marcy A., David S. Bauman, and Stephen W. Chapel, "Nuclear Construction Lead Times: Analysis of Past Trends and Outlook for the Future." The Energy Journal, January 1985, 45-62.

10. TVA Office of Nuclear Power. U.S. Nuclear Plants Cost Per KW Report. Washington: TVA, September 1987.

11. Yates, Marshall, "Nuclear Energy: A Failed Promise or a Promising Future?" Public Utilities Fortnightly, April 1, 1990, 12-13.

12. Zimmerman, Martin B., "Learning Effects and the Commercialization of New Energy Technologies." Bell Journal of Economics, Autumn 1982, 297-310.