.
t .
u
r . r
,
1 ..
O . .
I
. .
. . . .. .
. . . .. . . . . . .- 2! 15.
. §‘§§:fi
, . . 3.35"... .
..
y . .
.. 1
. f . .
. 1 u . 1 .1. .. . . . . 1 .
. s . . . .1 .. . . .. . . : . . . , . . a}!.$.flu.iol§‘lr
. .. . .. . ., . .. . . . . . 1 11.}!!! it’lrplév. I:
. A A ,. n l . . . t . 7 1 .. . psgvli.gifimm£
1 . a tin},
. . .
. . . V
. . h .
.
,. . .
. v
.. . 1 . .
~ .. . .
. ,. .
1 .
._ n
r
, I . .. . . .
a . . .
. o . v
.
. .
. . . .,. . . ..
.
. . o.
. . y . .
. . 1 1.
. A v u
z . ‘ a . .
.. . . . :
. . A .
.. . 1.
_.
. r .
. A . .
u . y ,
. . .._ . . .
. . . . . . f o. . ., .
‘ .. . . . .. a .
1 . . . .. . v. 1 . .
. . . . .
.1 . .

 

“1... .. .. .. . . .7. .. .. Iv . 1
. . ~ ..r . » .1 . .12... a... .3315...»

 

MWMMMWWWW
[LSMWARV a,
Miriam, M"?

burs

l Uriiwiewsitly3

 

This is to certify that the

dissertation entitled

The Size-Reliability Trade-off
and the Construction Cost of ,
Coal-Burning Generating Units.

presented by

Bradley K. Borum

has been accepted towards fulfillment
of the requirements for

Ma]— degree in icnnnmics_

Major professor

/’-\ ‘
I
Date 0“" ‘7‘; “’80!

MSU is an Affirmative Action/Equal Opportunity Institution 0-12771

(n

a 4);: a: s /

’-

¢ ~5_-r--+

 

PLACE “ RETURN BOX to romwo this checkout from your record.
TO AVOID FINES mum on or boron duo duo.

DATE DUE DATE DUE DATE DUE
In 2 9 ma

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

T—U

MSU lo An Afflrmdlvo Action/Equal Opportunity Intuition
emu-nut

_.__....... w

 

 

l

”-1—...-

 

THE SIZE-RELIABILITY TRADEOFF
AND THE CONSTRUCTION COST
OF COAL-BURNING GENERATING UNITS

By
Bradley Keith Borum

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Economics

1989

g) ABSTRACT
THE SIZE-RELIABILITY TRADEOFF

AND THE CONSTRUCTION COST
OF COAL-BURNING GENERATING UNITS

by
Bradley Keith Borum

The construction cost of a baseload coal-fired generating unit
should depend, in general, on the engineering attributes of the unit.
Previous studies generally found that there are scale economies in
average construction costs over the entire range of observed unit sizes.
However, these studies failed to include reliability as an attribute or
characteristic of the units. Larger generating units tend to be
considerably less reliable than smaller units. Thus, an important
unanswered question is whether large generating units have lower average
construction costs per Kw of capacity because of economies of scale or
because of poor quality. The implication is that omitting unit
reliability will bias the estimates of the construction cost function.
To analyze this question, we develop an ex ante long-run unit
reliability and construction cost model. Both unit reliability and
construction cost were expressed as functions of the dominant design
characteristics of the generating unit. Furthermore, a simultaneous
relationship between construction cost and reliability was assumed. The
simultaneous-equation model was estimated with a data set that contained
observations on 84 coal-fired units that entered commercial operation

during the period 1964-1974. The regression estimates were used to

evaluate how capital costs per KH and capital costs per KNH responded to

 

Bradley K. Borum

variations in unit size and reliability. Both measures of capital costs
were found to be characterized by economies of scale at low levels of
reliability and diseconomies at high levels of reliability. The cost
minimizing level of reliability for each measure was also found to
decrease as unit size increased. Capital cost per KWH generated are
found to be lowest for units in the 300-400 MN range with equivalent
availabilities of 85 to 90 percent.

Copyright by
BRADLEY KEITH BORUM
1989

ACKNOWLEDGMENTS

Special thanks go to Dr. Harry M. Trebing for stirring my interest
in the field of public utility economics and his continued encouragement
to complete this project. The other members of my dissertation
committee, Drs. Ken Boyer, Warren Samuels and Peter Schmidt, were
unusually helpful and prompt with their suggestions. It is no
exaggeration to say that their suggestions greatly improved the entire
study.

Special thanks also go to 8.". Stoller Corporation. This study
would not have been possible without the unit availability data it so
kindly provided to me. Kesa Hall, Beth Hynes and Joyce Smith typed
various parts of this manuscript and their assistance is greatly
appreciated.

Finally, special recognition must go to Julie Relue. This project
might never have been completed without her kindness, understanding and

moral support.

TABLE OF CONTENTS

xi

List of Tables ...................... viii
List of Figures ......................
CHAPTER 1: INTRODUCTION ................. l
The Cost of Poor Unit Reliability .......... 2
Relationship Between Unit Attributes and Construction
Costs ........................ 5
The Changing Electric Utility Environment ...... 7
Methodology ..................... 8
CHAPTER 2: SURVEY AND CRITIQUE OF THE LITERATURE ..... 11

The Basic Production Process and Technological Change 11

Models Which Fail to Properly Control for Capital

Heterogeneity .................... 14
Models Which Avoid the Vintage Problem ........ 25
Summary ....................... 43
CHAPTER 3: DEVELOPMENT OF A GENERATING UNIT CONSTRUCTION

COST AND RELIABILITY MODEL .......... 46
Introduction ..................... 46

The Heterogeneous Nature of Capital ......... 47

The Two-Dimensional Nature of Output ...... 47

The Unit Design Process and Ex Post Production . 48

The Model ...................... 51
General ..................... 51
The Expected Ex Post Cost Function ....... 52

vi

Ex Ante Costs and the Investment Decision . . . . 53

Specification of the Construction Cost and
Reliability Model .................. 55

The Determinants of Unit Construction Cost . . . 55
The Determinants of Unit Reliability ...... 59
The Need for a Simultaneous-Equations Model . . . 63

Estimation of the Simultaneous-Equations Model . 66

Empirical Analysis .................. 67
The Data .................... 67
Regression Results ............... 69
Summary ....................... 97
CHAPTER 4: AVERAGE CAPITAL COSTS AND THE UNIT SIZE-
RELIABILITY TRADEOFF ............. 100
Introduction ..................... 100
The Behavior of Average Capital Costs ........ 101
Efforts to Improve Utility Performance and the Size-
Reliability Tradeoff ................. 120
Summary ....................... 124
CHAPTER 5: CONCLUSIONS .................. 126
Summary ....................... 126
Suggestions for Future Research ........... 130
APPENDIX A: THE DEFLATION PROCESS ............ 132
APPENDIX B: ESTIMATION OF A COBB-DOUGLAS SPECIFICATION OF
THE COST FUNCTION .............. 133
APPENDIX C: DERIVATION AND STATISTICAL SIGNIFICANCE OF
NONLINEAR DERIVATIVES ............ 141
END NOTES ........................ 143
BIBLIOGRAPHY ....................... 149

vii

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

1-1:

2-1:

2-2:

3-1:

3-2:

3-3:

3-4:

3-5:

3-6:

3-7:

3-8:

3-9:

3-10:

3-11:

LIST OF TABLES

Average Equivalent Availability Factors in the

United States, 1971-1980 ........... 3
Capacity Additions by Technological Group and
Year: 1950-1982 (% of New Capacity) ..... 13
Size Distribution of New Coal Capacity

Year: 1950-1982 (Mwe) ............ 14
Ordinary Least Squares Estimate of the
Construction Cost Function .......... 72
Ordinary Least Squares Estimate of the
Construction Cost Function .......... 73
Ordinary Least Squares Estimate of the
Construction Cost Function .......... 74
Ordinary Least Squares Estimate of the
Construction Cost Function .......... 75
Ordinary Least Squares Estimate of the

Reliability Function ............. 77
Two-Stage Least Squares Estimate of the
Construction Cost Function .......... 78
Two-Stage Least Squares Estimate of the
Construction Cost Function .......... 79
Two-Stage Least Squares Estimate of the
Construction Cost Function .......... 80
Two-Stage Least Squares Estimate of the
Construction Cost Function .......... 81
Two-Stage Least Squares Estimate of the
Reliability Function ............. 82
Plant Cost Elasticities with Respect to

Size for a Generic Unit ........... 84

viii

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table
Table
Table
Table
Table
Table
Table
Table
Table
Table

3-12:

3-13:

3-14:

3-15:

3-16:

3-17:

3-18:

3-19:

3-20:

3-21:

3-22:

4-1:
4-2:
4-3:
4-4:
4-5:
4-6:
4-7:
4-8:
4-9:
4-10:

Plant Cost Elasticities with Respect to
Size for a Subcritical Unit ......... 85
Plant Cost Elasticities with Respect to
Size for a Supercritical Unit ........ 86
Plant Cost Elasticities with Respect to
Size for a Generic Unit ........... 87
Plant Cost Elasticities with Respect to
Size for a Subcritical Unit ......... 88
Plant Cost Elasticities with Respect to
Size for a Supercritical Unit ........ 89
Plant Cost Elasticities with Respect to
Reliability for a Generic Unit ........ 90
Plant Cost Elasticities with Respect to
Reliability for a Generic Unit ........ 91
Plant Cost Elasticities with Respect to
Reliability for a Subcritical Unit ...... 92
Plant Cost Elasticities with Respect to
Reliability for a Supercritical Unit . . . 93
Plant Cost Elasticities with Respect to
Reliability for a Subcritical Unit ...... 94
Plant Cost Elasticities with Respect to
Reliability for a Supercritical Unit ..... 95
Capital Cost Per KWH for a Subcritical Unit . . 103
Capital Cost Per KW for a Subcritical Unit . 104
Capital Cost Per KWH for a Subcritical Unit . . 105
Capital Cost Per KW for a Subcritical Unit . . 106
Capital Cost Per KWH for a Supercritical Unit . 107
Capital Cost Per KW for a Supercritical Unit . 108
Capital Cost Per KWH for a Subcritical Unit . . 109
Capital Cost Per KW for a Subcritical Unit . 110
Capital Cost Per KWH for a Supercritical Unit . 111
Capital Cost Per KW for a Supercritical Unit . 112

ix

Table
Table
Table
Table
Table

Table

Table

Table

Table

4-11:
4-12:
4-13:
4-14:

8-2:

B-3:

B-4:

B-5:

Capital Cost Per KWH for a Subcritical Unit . 113
Capital Cost Per KW for a Subcritical Unit . . 114
Capital Cost Per KWH for a Supercritical Unit 115
Capital Cost Per KW for a Supercritical Unit . 116

Ordinary Least Squares Estimate of the
Construction Cost Function .......... 134

Ordinary Least Squares Estimate of the
Construction Cost Function .......... 136

Two-Stage Least Squares Estimate of the
Construction Cost Function .......... 137

Ordinary Least Squares Estimate of the
Construction Cost Function .......... 139

Ordinary Least Squares Estimate of the
Construction Cost Function .......... 140

LIST OF FIGURES

Figure 3-1: Possible Forms for the Incremental Heat

Rate ..................... 50
Figure 4-1: Capital Cost Per KW ............. 118
Figure 4-2: Capital Cost per KWH ............. 119

xi

CHAPTER 1

INTRODUCTION

In this dissertation we show that the existence of economies of
scale with respect to the construction costs of coal-fired generating
units depends on the interaction of the size and reliability of the
unit. The costs of building coal-fired units is characterized by
economies of scale at low levels of unit reliability, constant returns
at higher levels of reliability, and diseconomies at very high levels of
reliability. This result is in sharp contrast to the conclusions
reached by all other researchers and is of interest for a number of
other reasons. First, our results are derived by explicitly taking into
account that reliability, as measured by equivalent availability,‘ is an
attribute of coal-fired units and in general is inversely related to the
size of the unit. Second, we recognize the simultaneous nature of the
choice of the attributes of a generation unit and its construction cost.
Third, the Federal Energy Regulatory Commission (FERC) has proposed
changes in policy which could substantially deregulate the generation of
electricity. The proposals deal with competitive bidding for new
generation capacity, independent power producers, and the administrative
determination of avoided costs. A major concern under all-source
bidding is how to evaluate and weigh the price and nonprice
characteristics of each proposal when ranking the bids. An important
nonprice attribute to include when evaluating a bid for baseload
generation capacity is the reliability or availability of the unit.

Each of these reasons will be examined in greater detail below.

1

The Cost of Poor Unit Reliability

A general result of previous studies is that there are scale
economies in construction costs over the entire range of observed unit
sizes.2 These studies ignore the ex ante design process in which
reliability is a key parameter. As a result, these studies overlook the
possibility that a relatively inexpensive baseload coal-fired unit to
build may not be so inexpensive to operate if it is frequently out of
service due to forced outages or necessary maintenance.

The key issue here is the intensity with which a unit is used.
Baseload units are meant to be operated at maximum capacity '
continuously. The capital intensive nature of baseload coal-fired
generation means that the average total generation costs of a unit can
be significantly reduced if the unit is used intensively.

Electric utilities consistently increased the average size of new
generating units until the mid-1970’s.3 This trend was based on a
number of widely held propositions: construction costs per kilowatt of
capacity fell as unit size increased;‘ operation and maintenance costs
per kilowatt-hour (kwh) could be reduced by building larger units; and
that new, larger units historically had lower heat rates.

In general, however, there is a negative relationship between the
size and reliability of a generating unit (see Table 1-1). There are a
number of reasons for this inverse relationship. One is that the
movement to larger units was accompanied by a movement to higher steam
pressure conditions. Previous studies generally have found that
increases in steam pressure conditions have been associated with higher

forced outage rates.5 Second, similar types of outages are generally

3

Table 1-1. Average Equivalent Availability Factors
in the United States, 1971-1980

 

 

 

Average Equivalent
Iechnglogy - Unit Qaptgitx Attilabiltty Fgctor
All fossil-fueled steam 79.61
100 - 199 MW 83.05
200 - 299 MW 80.32
300 - 399 MW 73.88
400 - 599 MW 72.02
600 - 799 MW 69.33
800 MW and above 68.49

 

 

 

Source: Joskow and Schmalensee, 1983

4
longer for larger units than for smaller units. This is due to the
larger area and parts that have to be repaired. And third, more
materials and equipment simply create more opportunities for breakdown.6

The reliability of a utility system is defined as the ability of
the system to meet the demand for power at any given point in time.
Individual generating unit reliability is one of the primary
determinants of system reliability. The more frequent a generating unit
breaks down, the lower the reliability of the system. Thus, the lower
average reliability of larger units adds to a utility’s total system
costs because additional capacity must be built if a given level of
system reliability is to be maintained.

Our primary concern is with total per-kwh generation costs at the
level of the individual unit. The annual total generation costs of a
unit consist of fuel expenses, capital-related costs, and labor
expenses. In general these account for 50%, 40%, and 10%. respectively,
of total generation costs. Annual capital-related costs are by
definition fixed and are the product of the total construction costs of
the unit and the annual fixed charge rate for the utility. The fixed
charge rate basically consists of property taxes assessed on the unit
and the cost of stocks and bonds issued by the utility to finance
construction of the unit. A poor level of reliability means that the
capital-related costs of a generating unit are spread over a lower level
of output than if the unit is used more intensively. Thus, poor unit

reliability results in higher average generation costs for the unit.

5

Poor reliability can also reduce the thermal efficiency of a
generating unit.7 Good heat rate performance is related in general to a
high level of utilization because heat loss is fairly constant at any
load. So heat loss is relatively larger at low load than at a high
load. Deratings also increase heat rates, because restarting a unit
means that heat energy must be expended to reheat the boiler and other
components. As a result, frequent deratings of a unit due to outages
can be expected to increase the unit’s average fuel costs. The heat
rates of large units are more likely to be adversely affected by
frequent deratings because large units, in general, are more prone to

outages.

e t o h t n ribut u i n Co

The decision to build a baseload coal-fired unit means that the
utility is purchasing a piece of capital equipment with various
engineering attributes. Therefore, when comparing different generating
units (or any type of capital equipment for that matter), one should
make adjustments for any differences in key attributes of the units.
The basic concept is that the cost of a generating unit should depend on
the attributes of the unit. A number of key design decisions that
significantly affect the costs of building a baseload coal-fired unit
have been identified in the engineering and economic literature.8 These
are: the size of the unit, unit order and replication, coal type, steam
pressure conditions (subcritical or supercritical), pollution control

strategies, cooling method, and reliability.

6

Previous studies generally found that there are scale economies in
construction costs over the entire range of observed unit sizes.
However, these studies failed to include reliability as an attribute or
characteristic of the units. Larger generating units tend to be
considerably less reliable than smaller units. So one has to wonder
whether large generating units have lower average construction costs per
kw of capacity because of economies of scale or because of poor quality.
The implication is that omitting unit reliability will bias the
estimates of the construction cost function.

A unit's attributes and construction costs are determined
simultaneously. A utility might be expected to modify its choice of
unit attributes if it knows something about the error term it faces in
the construction cost function. This means that the choice of
attributes will be correlated with the error term. This will result in
ordinary least squares (OLS) estimates of the cost function being
biased. (OLS is the most widely used estimation technique in previous
studies.) If a complete model involves the simultaneous determination
of attributes and cost, then simultaneous estimation techniques are
necessary.

Note that simultaneous equation error will occur only if the
utility has some conception of the error term it faces. But it would
seem to be unreasonable to assume that the utility is totally unaware of
the direction and size of the error. After all, each major utility in

the nation has considerable experience participating in the construction

7
of different types of generating units. Also, a utility that is
considering building a generating unit can draw on the experience of

other utilities that have recently built similar types of units.

the Changing Elggtrjg Utjljty Entjnnnngnt

Three controversial notices of proposed rule makings (NOPRs) were
recently issued by the FERC. The NOPRs deal with competitve bidding for
new generation capacity, independent power producers (IPPs), and the
administrative determination of avoided costs. The most controversial
aspect of the NOPRs is that they would include IPPs as potential sources
of new generation capacity under any competitive bidding programs.

A major concern under any all-source bidding program is, how to
evaluate and weigh the price and nonprice characteristics of each
proposal when ranking the bids.9 An important attribute to include when
evaluating a bid for baseload generation capacity is the reliability or
availability of the unit. Baseload units are meant to be operated at
maximum capacity whenever they are available. Each generating unit
design is likely to have a different level of reliability. Also, each
design can enhance reliability by adding redundancy to key components
and/or by specifying the use of more reliable but more costly materials
and equipment. Thus, it is extremely important to determine the
consistency between the design of the unit, its construction cost, and
its reliability.

Numerous states also have implemented or proposed performance

0

standards for individual generating units.‘ The two most common

criteria used to measure unit performance are equivalent availability

8
and heat rate. Utilities operating units that are found to be efficient
are financially rewarded, while utilities with units that fail to meet
specified minimum performance levels are penalized. One possible
problem is that imposing minimum standards for a narrow range of
performance criteria can create perverse incentives for the utility.
For example, the utility might be willing to incur excessive costs in
areas outside the incentive program so as to improve performance in the
targeted areas. In this context, it is important to understand the
extent to which there is a tradeoff between the reliability and the
construction costs of a generating unit. The existence of such a
tradeoff would call for regulatory policies focused on a utility’s
design and construction decisions. If, on the other hand, unit
performance is a random variable or a function of utility maintenance

policies, then different regulatory policies would be necessary.

nethodolggy

When comparing pieces of capital equipment it is necessary to
control for all relevent engineering characteristics. Most previous
studies have treated the capital embodied in generating units as if it
was homogeneous. Other studies have disaggregated the attributes of
generating units along the dimensions of unit size and steam pressure
conditions (or heat rate). These studies ignored the ex ante design and
construction process in which reliability is a key parameter. As a
result, these studies overlooked the possibility that a relatively
inexpensive coal-fired unit to build may not be so inexpensive to

operate if it is frequently forced out of service due to mechanical

9

failures. This is especially relevant given the capital intensive
nature of baseload electric power generation which means that average
total generation costs can be significantly reduced through intensive
utilization of the unit.

The central feature of the model developed here is that the cost of
a generating unit depends on the engineering characteristics of the
unit. Thus, emphasis will be placed on a multidimensional description
of capital equipment that includes reliability as an attribute. We
argue in Chapter 3 that the traditional neoclassical production model is
inappropriate when capital has multiple attributes. Therefore, we have
developed an engineering cost function that is flexible enough to allow
all major engineering characteristics to be included.

A critical review of the literature is presented in Chapter 2.
This is followed by a presentation of the theoretical basis for the
structure of our model in Chapter 3. There we argue that minimizing a
unit's annual total per-kwh generation costs can be approximated by
minimizing its annual capital cost per-kwh. This is based on the
concept that intensive utilization of a unit not only spreads its annual
capital-related costs over a higher level of output, but average per-kwh
fuel costs are also reduced. We have developed an engineering
construction cost function which includes reliability as one of the
characteristics of a generating unit. This model enabled us to
determine the cost minimizing combination of unit size and reliability
while accounting for other important engineering attributes.

Chapter 3 also includes a specification of the construction cost

model and a discussion of error structure and estimation technique. It

10
concludes with econometric estimates of the construction cost function
and an examination of the elasticity of per-kw construction cost with
respect to unit reliability. In Chapter 4, we show (assuming that
whenever a unit is available it is used) how annual capital costs
per-kwh change for various combinations of unit size and reliability.
Chapter 5 provides a summary of our results and ideas for further

research.

CHAPTER 2

SURVEY AND CRITIQUE OF THE LITERATURE

The electric generating industry has been the subject of a large
number of econometric studies of its production process. This is a
result of the capital intensive nature of the generation process, the
rapid technological advancement embodied in the generation equipment,
and the abundance of detailed data at the plant and firm levels due to
federal and state regulation of the electric utility industry.

This review concentrates on that part of the literature which
examines the capital cost of steam electric generating units. Emphasis
is placed on the methodologies and data used in the studies that are
reviewed. Particular attention is placed on the fact that few studies
have properly addressed the heterogeneity of capital equipment and its
role as a channel for technological advancement in the electric

generation process.

The Basic Prodnctjgn Procgtt and Igghnnlngigal Change

Electricity is generated by a process that involves transforming

 

energy from one physical state to another. The technology for
transforming fossil fuel into electricity using steam turbines is well
developed. Fossil fuel is burned in a furnace to generate heat.
Pressurized high-temperature steam is created by transferring the heat
to water circulating within an enclosed boiler. The pressurized steam
is expanded through a turbine which turns a generator to produce

electricity. The steam is then cooled in the condenser and returned to

11

12
the boiler to repeat the cycle. Thus, the energy transformation process
can be divided into four stages--fuel combustion, steam generation,
steam expansion, and power generation."

Each stage of the production process is represented by a different
piece of capital equipment. It is here that capital heterogeneity is
introduced. The furnace is linked to combustion, the boiler to the
creation of steam, the turbine to steam expansion, and the generator to
the production of electricity. Each piece of capital equipment is
purchased with numerous engineering characteristics which are
substituted for one another at the design stage of the construction
project. This causes heterogeneity at the individual component level,
but heterogeneity is also created at the generating unit level by the
physical linking of the components. As a result, a generating unit
"represents a series of individual and joint optimizations that are
reflected in the designs” of the individual components and the unit as a
whole.12

Another source of capital heterogeneity is the innovation process
itself. Traditionally the manufacturers of electrical equipment have
played a key role in the development of technological innovations in
electric power generation. This means that the vast majority of
technological innovations have been capital-embodied."3 Electric
utilities participated in the innovation process by being among the
first to purchase the equipment and use it under actual operating

conditions. As the innovation proves itself technically and

l3
economically it is gradually implemented by other utilities. As a
result, a number of technologies can and do coexist at any given time in
the industry.
It is also important to note that technological innovation and

‘ The primary

average unit size are positively correlated over time.1
goal of technological innovation has been to improve the thermal
efficiency of the production process. The desire for increased thermal
efficiency led to continued efforts to increase steam pressure and
temperature conditions. Table 2-1 shows the steam pressure conditions
of all coal-fired units placed into commercial operation between 1950
and 1982. This table shows that the movement to higher pressure (more
technologically advanced) units occurred only gradually. Table 2-2
shows the size distribution of the corresponding units. Table 2-2 shows
that average unit size increased rapidly until the early 1970’s.
Together the tables show that the movement to higher pressure (more
technologically advanced) units was accompanied by a movement to larger
sized units.

Table 2-1. Capacity Additions by Technological Group and
Year: 1950-1982 (% of New Capacity)

 

 

Period Turbine Throttle Pressure Groups (psi)
l§QQ_Qr_Le;§ 1899 2999 2999 3599
1950-1954 39 45 13 2 0
1955-1959 10 32 36 20 1
1960-1964 2 21 20 45 12
1965-1969 2 8 1 46 42
1970-1974 0+ 5 0+ 32 62
1975-1980 0 6 1 62 31
1981-1982 0 2 4 88 6

 

 

 

Source: Joskow and Rose, 1985

14

Table 2-2. Size Distribution of New Coal Capacity
Year: 1950-1982 (Mwe)

 

 

Number of New
Period Mean Minimum Maximum Units Installed
1950-1954 124 100 175 99
1955-1959 168 100 335 175
1960-1964 242 100 704 104
1965-1969 407 103 950 100
1970-1974 591 115 1300 109
1975-1980 545 114 1300 127
1981-1982 517 110 891 41

 

 

 

Source: Joskow and Rose, 1985

The main point is that the use of vintages to define periods of
technologically homogeneous capital in the electric generation industry
is likely to be inadequate. Tables 2-1 and 2-2 indicate that there are
a number of technologies in use at any point in time, and that unit size
and advanced technology are closely related. Thus, failure to
adequately control for technology and size is likely to bias any
parameter estimates.

As a result, the studies reviewed here are classified into two
groups--those studies that explicitly take into account or somehow
control for capital heterogeneity when estimating the construction cost

function parameters and those studies that do not.

0:: .1 .1 F. . ' o--r '1 - . .-i al -te --:neit
Komiya (1962) studied the ex ante production function for steam
electric generation and sought to explain shifts in the production

function due to technological change. His sample consisted of 235 new

15
plants built between 1938 and 1956. The plants included both single and
multiple units. The sample was divided into eight vintage-fuel type
groups. Each fuel type, coal and noncoal, was divided into four
technological vintage periods: 1938-45, 1946-50, 1951-53, and 1954-56.
The idea was that each plant that entered commercial operation in a
particular period embodied the best technology available at the time.
Any differences in the production function across vintages would be
evidence of technological change.

Komiya estimated a Cobb-Douglas production function within each
cell, but he concluded that the technology did not allow input
substitution when the function performed poorly. He then estimated a
Leontief type, or fixed proportion, model within each cell.

YF - AF + BFX,

Yc - Ac + BCX, + BNX2

YL - AL + BLX1 + BNX2
Where,

YF - natural log of fuel input, total BTU’s, per generating unit

when operated at full capacity,

Yc - natural log of average equipment cost per generating unit in

constant dollars (Note that the cost of structures and land were

excluded.),

YL - natural log of the annual average number of employees per

generating unit,

X, - natural log of the average size of the generating unit in

megawatts,

X2 - natural log of the number of generating units in the plant.

l6

Komiya found that economies of scale at both the unit and plant
levels were important factors in declining input requirements over time.
Scale elasticities at the plant level for fuel and capital were
estimated to range between .80 and .85. The scale elasticity with
respect to labor was estimated to range between .50 and .60. Technical
change was found to have little impact on capital or fuel input
requirements. The effect of technological change was to reduce labor
input by 46 percent from vintage 1938-45 to vintage 1954-56 for coal
plants of a given size. There was also a significant difference in
capital equipment requirements of equal sized coal and noncoal plants.

The major weakness in Komiya’s study is the use of vintage cells to

5 As noted earlier, a

characterize periods of homogeneous technology.1
number of technologies coexist at any given time in the industry. Also,
improvements in technology are associated with increases in average unit
size. This could cause his capacity parameter estimates to be biased.
Barzel (1964) estimated log-linear input demand functions for fuel,
labor, and capital. His sample consisted of 220 plants that entered
commercial operation between 1941 and 1959. Each plant’s annual data
was observed from its first full year of operation until 1960 or until

there was a major change in the plant.

The capital input demand equation is:
4 18
1'09 Pk I iE‘I bill 09x. 41-h}; b'x'

Where,

PK - total undeflated value of plant including equipment,
structure, and land,

X, - plant size measured in kilowatts,

17
X.2 - labor price observed in first full year of operation,
X, - fuel price observed in first full year of operation,
X, - plant factor in first full year of operation,

X5“, - vintage dumies.

Barzel includes the ex post plant factor as a proxy for the desired
rate of utilization of the plant. He appears to be the first to
recognize that plants with higher levels of desired utilization must
have components which can handle the added stress, thus requiring more
capital investment.16

Barzel finds that the coefficient of the size variable is .815. He
concludes that there are economies of scale in the capital cost of the
plant since the coefficient is significantly smaller than unity. He
also estimated that the elasticity of plant investment with respect to
the plant factor was .117. As a result, he concludes that quality is an
important determinant of the costs of capital equipment.

A major shortcoming of Barzels’ study is the use of vintage dummy
variables. The dummy variables were included to shift the intercept
over time in response to technological change, but there are at least
two problems with this methodology. First, the shifts are partly due to
inflation since Barzel used the total undeflated value of the plant as
the dependent variable.17 Second, the period of time covered by his
sample was one of considerable technological change and, given the
deliberate pace of innovation in the industry, there is likely to be a

number of technologies in use at any given time. Again, technological

18
change and the size of plants or units is correlated so that poor
treatment of technological change can seriously bias the econometric
results.

Another problem with Barzels’ study is that the ex post plant
factor observed in the first full year of operation is likely to be a
poor proxy for the level of desired utilization for two reasons. First,
availability and unit size are inversely related, while unit size and
desired utilization generally are positively correlated. Thus, the ex
post plant factor may understate the level of desired utilization due to

° Second, generating units

the declining availability of larger units.1
may go through a break-in period the first year or two of commercial
operation. The break-in period may be characterized by high forced-
outage rates and derating or cycling of units meant for baseload
operation."

Galatin (1968) estimated input requirement functions for fuel,
labor, and capital. His sample included 158 plants which entered
commercial operation from 1920 to 1953. Only plants with units of the
same vintage and size were included in the sample. The sample was
divided into 12 vintage-fuel subsamples so that the effects of scale and
technological change could be examined across the vintage cells. The
six vintages were 1920-24, 1925-29, 1930-39, 1940-44, 1945-50, and
1951-53.

Galatin postulated the capital cost of a generating unit is a
function of the size of the unit and the number of units in the plant.

The functional form was specified as:

CT/N or CE/N - A,N + A,"2 + A3N3 + A,x,<

19
Where,

CT - total undeflated capital cost of a plant including land,
structures, and equipment,

CE - total undeflated equipment cost of a plant,

N - number of units in the plant,

XK - size of each unit in megawatts.

The capital cost function was estimated with data covering the
vintage fuel-type cells 1945-50 and 1951-53, so that costs for land,
structures, and equipment ”may be assumed to relate to approximately the
same period."20 Galatin found that total capital cost and equipment
cost per generating unit increased with the size of the unit. He also
found that total capital cost and equipment cost per generating unit for
coal-fired plants in the 1945-50 vintage fell as the number of units in
the plant increased from one to three units, and increased as the number
of units at the plant increased beyond three. Average costs per
generating unit for mixed fuel-type plants of 1945-50 vintage and
noncoal plants of 1951-53 vintage also fell as the number of units in
the plant increased from one to two units, and then rose as the number
of units increased beyond two.21

There are a number of problems with Galatin’s study. One is that
the analysis of capital input is only applicable to plants composed of
units of the same size, vintage, and fuel type. Secondly, the sample
includes plants with units ranging in size from 4 MW to 150 MW.

However, Galatin fails to recognize that desired utilization generally
increases with unit size, and that the desired level of utilization,

independent of unit size, is an important determinant of plant capital

20
costs.‘22 Thirdly, Galatin used vintages in an effort to control for the
effects of changing technology.

The primary objective of Huettner’s study (1974) is to examine the
effects of technological change and plant capacity on the average
investment required per unit of capacity. Huettner’s sample consisted
of 391 plants divided into 13 vintage time periods from 1923-1968.
Within each of the vintage cells he estimated an average capacity cost
equation which included fuel type dummies and the reciprocal of the size
of the plant as the primary explanatory variables. Huettner used
stepwise regression analysis due to a conflict between the number of
explanatory variables and the number of observations within each vintage
cell.

Huettner did some preliminary testing of the model on a subsample
in order to reduce the number of explanatory variables to be considered
when using stepwise regression analysis for the entire sample. The
subsample consisted of 185 subcritical, coal-fired plants of full indoor
construction. The subsample ranged from 5 observations to 30
observations per vintage period and averaged 14 observations per period.
Due to the small number of observations in some of the vintage cells,
Huettner included only three explanatory variables in the average
capacity cost equation estimated with the subsample. The three
variables chosen were the number of units in the plant, the fuel
consumption of the plant, and the reciprocal of plant capacity.

Huettner included the fuel consumption variable, as measured by the
plants average heat rate, to test whether load type has an effect on

average capacity cost. He noted,

21

Generating plants may also be classified as base-load plants,

cycling plants, and peak-load plants. Base-load plants are

designed to operate at maximum fuel efficiency without being

shut down for long periods of time. This may increase unit

capacity costs (UCC) above that of cycling plants which are

designed to operate at the highest fuel efficiency consistent

with rapid warm-up and cool-down during frequent shutdowns.

Cycling plants might tend to have lower UCC due to looser

tolerances on equipment, as for example on the turbine

bl ades.”

Huettner found the fuel consumption variable was generally of the
appropriate sign, but statistically significant in only 2 of 13 vintage
periods. He noted there was a high degree of correlation between plant
heat rate and plant capacity which meant that multicollinearity was
likely to be a problem. As a result, Huettner excluded the fuel
consumption variable from the stepwise regression analysis of the full
sample.

The number of units variable was included by Huettner because "some
studies argue that quantity discounts are a significant factor affecting
equipment costs in generation plants: hence, one would expect UCC to
decline as the number of units increased.“" Huettner found that plants
with multiple units had lower average capacity costs in 10 of the 13
vintage periods, but the coefficient was statistically significant in
only 3 of these 10 periods. It should be noted that this variable was
dropped from the remainder of the analysis due to its poor performance.
As a result, Huettner’s study analyzes plant-level economies instead of
unit-level economies.

Huettner found that average capacity costs generally declined as
the size of the plant increased. The capacity variable coefficient was

positive in 9 of the 13 vintage periods and statistically significant in

7 of these periods. Based on the performance of the capacity variable,

22
it became the primary variable of interest in the stepwise regression
analysis of the full data set.

Plant characteristics used as explanatory variables in the capacity
cost equation for the full sample included fuel-type dummies, a
full-indoor construction dummy, a supercritical dummy, and the
reciprocal of plant capacity. Huettner found that plant fuel-type is a
significant determinant of the average capacity costs of generating
plants. Coal-fired plants were more expensive to build than oil-fired
plants, and oil-fired plants were more costly than gas-fired plants.
Supercritical plants were represented in only the 1959-60, 1963-65, and
1966-68 vintage periods, but were significantly more expensive than
subcritical plants in 2 of the 3 periods. Huettner also found that
average capacity cost declined with increased plant capacity in every
vintage period since 1940.

There are a number of problems with Huettner’s study. First is the
use of vintages to define periods of homogeneous technology. Second,
Huettner recognized that baseload and non-baseload plants have
significantly different design characteristics due to differences in
desired utilization, and that baseload Operation and plant size are
highly correlated. But the estimated plant size coefficients will be
biased by his failure to control for differences in desired utilization,
and the mixing of baseload and non-baseload plants in the sample.

Third, Huettner recognized that the reliability of a generating unit is
inversely related to its size and “that the lower reliability of large
units reduces their economic attractiveness.“” But he fails to include

reliability as an explanatory variable so the parameter estimates may be

23
biased. As a result, it is impossible to determine whether large units
cost less to build due to economies of scale or poorer quality as
reflected by lower reliability.

Wills (1978) recognized the non-homogenous nature of capital
equipment in the steam-electric generating industry and noted the
construction cost of a plant will be a function of its attributes.
Thus, Wills estimated an hedonic cost function for steam electric
plants. His sample included 156 plants which entered commercial
operation between 1947 and 1970. The observations were divided into
eight vintage cells.

The attributes initially considered by Wills were plant capacity,
average unit capacity, unit fuel efficiency, and the average number of
employees that worked in the plant. He also noted that plants could be
divided into groups based on construction-type, fuel-type, and whether
the plant contained a single unit or multiple units. However, Wills
concluded from a preliminary analysis of the sample that the effects of
the dummy variables could be restricted to an interaction with the
capacity variable. He also excluded the fuel efficiency variables
because they were collinear with the size variable.

As a result, the hedonic capital cost equation was of the following
form:

PRICE . a0 + (a, + BI + Y, + 5,, + Nc)Cap + ¢2CAP2
Where,

PRICE - total nominal cost of the plant,

CAP - plant capacity,

B.- one if the plant is of full-indoor construction, zero
otherwise;

24
n - one if the plant has only one unit, zero otherwise;
a, - one if the plant burns coal, zero otherwise;
1% - dummy variables that represent the vintage of the plant.
The model was normalized by plant size in order to remove a problem
with heteroskedasticity. The model became:

Price/Cazp - «0(1/Cap) + (a, + B, + YI + 5,, + N) (Cap/Cap)
+ «2(Cap /Cap)

Wills argued that all of the explanatory variables were exogenous
except the capacity variables. He believed it likely that plant size
was correlated with the error term. Thus, he used instruments for unit
size multiplied by the number of units in the plant as instruments for
plant size. The instruments for plant size included the "expected
absolute growth in demand for electricity from the utility times the
number of units, the price of fuel times the number of units, the price
of fuel squared times the number of units, and expected demand growth
times the price of fuel times the number of units.”3

The hedonic cost function was estimated using random components
instrumental variable estimation and random components estimation.
Wills found that economies of scale in plant cost per unit of capacity
are essentially exhausted at plant capacities of about 100 megawatts.
He also found that vintage effects were not important determinants of
plant cost per unit of capacity.

There are a number of problems with Wills’ study. Once again there
is the problem of using vintages to define periods of homogenous
technology. Also, there is the problem of mixing baseload and
non-baseload plants in the sample. The sample includes plants that

range in size from 5 MW to 950 MW. Unit size and desired utilization

25
intensity are correlated but each, independent of the other, is an
important determinant of the cost of building a generating unit. Thus,
failure to control for desired utilization and mixing baseload and

non-baseload plants in the sample will bias the coefficient estimates.

ModelLWhich Avoid titfi Yintagg Problem”

Stewart (1979) is concerned with the relative importance of
capacity utilization and size of plant in determining the average cost
of generating electricity. He adopts a quasi-engineering approach by
incorporating technical information on the characteristics of the
capital equipment and production process. A quasi-engineering approach
is used because the investment decision of the utility "will encompass
determining both the segment of total demand the new plant will serve
and the configuration of the new plant." The load increment for which
the new plant is designed and built is "defined by an instantaneous
rate" (the size of the new plant), K, and the number of yearly hours the
plant will produce at that rate (or duration).”’ Stewart defined the
expected cumulative output of the plant as:

0 - 8760bK
Where,

8760 - number of hours in a year,

b - expected plant factor,

K - capacity of the plant measured in megawatts.

Stewart assumed that the range of technology available to the

utility could be fully defined by the size and thermal efficiency of the

26
generating unit. 50 the average capacity cost (cost per KW) function
was written as:

P, - Pk(¢,k)

The average capacity cost of a plant was expected to increase at an
increasing rate with the fuel efficiency, a, of the plant. Stewart also
expected the average capacity cost of a plant to decrease over some
range of plant size. As a result, the utility was faced with the
problem of choosing the cost minimizing level of fuel efficiency for a
unit designed to meet an expected load increment defined by b and K.

The cost minimizing heat rate was given as:

a' - g(K,b,PF,r)

Where,

¢° - cost minimizing heat rate (BTU/kwh),

PF - price per BTU of fuel,

r - cost of capital.

Ex ante total generation costs are derived by substituting a' into
the following cost function.

TC°(K,b,PF,r) - g(K,b,PF,r)8760bKPF +

rPK(g(K,b,PF,r),K)K
Where the first part of the total cost function represents ex ante fuel
costs and the second part represents ex ante capital cost.

The estimated plant cost function was combined with the plant’s
size, load factor, and factor prices to compute the cost minimizing heat
rate and the average total cost per Kwh for each plant. Stewart found

that plant utilization intensity was the dominant factor in reducing

27
average generation costs, while plant size was found to have relatively
little impact.
Our primary attention will be on Stewart’s plant cost function. He

estimated a translog Specification of the cost of plant function:

ln PK - A + Y, ln(a - T.) + Y... (m. - m2
+ YK ln(K) + Y,<,((ln(K))2 + Ydln(K)ln (a - I.)

+2Y,X,+u
I

Where,

PK - nominal land and equipment cost per KW of the generating unit
(excluding structures),

- average heat rate of the unit (BTU’s/Kwh),
- asymptotic heat rate (6000 BTU/Kwh),

K - capacity of the unit (Kw),

X.- regional dummies and natural log of the number of units in a
given plant.

The plant cost function was estimated using a sample of 58 plants
which entered commercial operation between 1970 and 1971. The sample
included plants with single units or multiple identical units. The
sample consisted of 19 steam electric plants and 39 gas turbine plants.

Stewart estimated two forms of the cost-of-plant function. One
included only a dummy variable to differentiate between the two types of
generating plants. Thus, the coefficients of the size and heat rate
variables were restricted to be equal for each type of plant. The
second specification allowed the plant type dummy variable to interact

with the size and heat rate variables and their interaction variable.

28
The plant type dummy was not allowed to interact with the squared size
or squared heat rate variables.

Stewart found plant equipment cost fell at a decreasing rate as
heat rate increased (thermal efficiency decreased) for both types of
plants. He also found plant size had very little impact on the cost of
equipment for either type of unit. Plant costs of gas turbines at the
mean heat rate declined with unit size only for units smaller than 70
MW. Average equipment cost for steam plants was found to increase at a
”relatively modest rate” over most of the reasonable range of unit sizes
and fuel efficiencies.

We believe there are a number of flaws in Stewart’s study. One
involves the estimation of a single quadratic function to approximate
both technologies. Steam-electric and gas-turbine technologies are
quite different so there is little reason to believe that scale
economies will be at all similar for the two technologiesf” A second
problem involves the mixing of baseload and non-baseload plants in the
sample. The steam electric plants range in size from 200 MW to 800 MW
and the gas turbines range from 20 MW to 187 MW. Failure to control for
different levels of desired utilization means the size coefficient will
be biased.

Another problem relates to Stewart’s failure to include reliability
as an attribute of a generating unit in the cost of plant function. He
is interested in the relative importance of capacity utilization and
plant size in determining the cost of generating electricity. However,
he fails to recognize that utilization depends on availability and that
availability is an attribute of a unit.

29

Komanoff (1982) estimated a log linear average construction cost
equation. The sample consisted of all U.S. coal-fired units, 100
megawatts or larger, which entered commercial operation from January
1972 through December 1977. The units ranged in size from 114 MW to
1300 MW and averaged 608 MW. Fifteen of the units had flue gas
desulfurization devices or scrubbers.

Real capital costs per kilowatt, excluding AFUDC, was regressed on
a number of explanatory variables using ordinary least square (OLS).
Unfortunately, Komanoff shows the econometric results for only the
statistically significant variables. All other variables were excluded
from the final regression equation and were simply listed as having been
tried in alternative specifications of the regression equation.
Insignificant variables included the presence of cooling towers,
supercritical boilers, and unit size.

Komanoff found the presence of scrubbers added 26 percent to the
capital cost of a generating unit. He also found that units which share
a plant site with an identical unit have lower capital costs than
non-multiple units. Capital costs were found to increase approximately
4 percent for each later year that the unit entered commercial
operation. Other significant explanatory variables were regional
dummies, which probably reflected regional variations in construction-
type, and the cost of labor. Units built in the Midwest or Northeast
are more likely to be of full-indoor design than units built in the
Southwest or Southeast.

Interestingly, Komanoff found capital costs to be significantly
correlated with ownership by two large utility holding companies: the

30
Southern Company and American Electric Power (AEP). Generating units
built by AEP were found to be 18 percent more costly than other
comparable units, while Southern Company’s units were 15 percent less
costly to build than other Southeast units. Komanoff argued these
differences were probably due to the two utilities’ unit design and
operation philosophies. AEP has a reputation for building highly
reliable and efficient units, while Southern Company tends to build
units with below-average reliability and fuel efficiency. As a result,
Komanoff concluded unit reliability is a significant determinant of a
units’ capacity costs.30

One problem with Komonoff’s study has to do with mixing baseload
and non-baseload units in the sample. That this may be a problem is
indicated by the inclusion of units as small as 114MW in the sample.
Baseload and non-baseload units have different levels of desired
utilization and, thus, different engineering characteristics which are
likely to effect the capital costs of each type of unit. Failure to
account for desired utilization can seriously bias the parameter
estimates since unit size and desired utilization are generally
correlated.

Another problem arises from Komanoff’s failure to properly account
for unit reliability in the capital cost equation. It is necessary to
control for different levels of desired utilization and to include a
reliability variable. This prevents the size coefficient from being
biased since size and reliability are inversely related. Komanoff also
implicitly recognizes the simultaneous nature of the relationship

between unit reliability and capital costs. He notes that higher

31
initial capital cost due to conservative design philosophy and, thus,
higher unit reliability may result in lower total costs over time. The
endogenous nature of unit reliability means simultaneous estimation
techniques are necessary for unbiased parameter estimates.

Perl (1982) is concerned with calculating the levelized cost of
electricity from coal-fired plants. ”Levelized costs are a constant
annual charge for electricity which yield the same present value as
actual annual charges over the life of a plant."31 Current accounting
practices and the capital intensive nature of coal-fired plants causes
high front end costs. Levelized costs, he argues, better reflect life
time electricity costs.

Perl’s methodology involves econometrically estimating capital
costs, non-fuel operating and maintenance costs, availability factors,
and heat rates for coal-fired units. These were regressed on the
engineering characteristics of the units using OLS. His sample included
245 coal-fired units which entered commercial operation between 1965 and
1980. These components of cost and performance were combined in the
following model to produce the levelized cost of electricity:

I RR.(1+ a)' 1

'3‘ *

I: G.(1 + a)'/(1 + r)' (1 + r)"'

i=1

Where,
RR- revenue requirement in year i,
G.- generation in year i,
N - book life of the plant,

M - number of years from current date to start of commercial
operation,

32

c - nominal discount rate,

r - inflation rate.

Perl assumed that plant life was 30 years, 1985 to 2014. He also
assumed when forecasting capital and O & M expenses per kilowatt-hour
that a unit is used if it is available.

We will focus our attention on Perl’s average capital cost and
equivalent availability equations. The dependent variable in the
equivalent availability equation is the logit transformation of the
equivalent availability factor. This transformation restricts the
estimated variable to the interval from zero to one. The sample used to
estimate the equation consisted of annual observations of equivalent
availability for ”a large sample of coal units operating from 1969
through 1977.“” Observations for the first full year of commercial
operation of a unit were excluded to avoid any bias due to break-in
problems. Explanatory variables included the year the unit entered
commercial operation to represent vintage, the age of the unit, the
reciprocal of unit size, and numerous dummy variables. The dummies
indicated a supercritical boiler, a balanced draft boiler, a cyclone
boiler, boiler manufacturer, turbogenerator manufacturer, and whether
the unit was built by the particular architect-engineer (A-E) or utility
represented.

Perl found equivalent availability tended to decline with increased
age and larger unit size. Subcritical units were also substantially
more reliable than comparable supercritical units. Perl also noted that
there appeared to be significant differences in the availability of
units built by particular A-Es and utilities.

33

Perl estimated a log linear specification of the capital cost
equation. The dependent variable is the natural log of real capital
cost per kilowatt excluding AFUDC. The sample consisted of 245
coal-fired units built between 1965 and 1980. Explanatory variables
included unit size, regional wages, the date the unit entered commercial
operation, and dummy variables indicating supercritical boilers,
scrubbers, and whether the unit was designed and built by a particular
A-E or utility.

Perl found average capital costs were significantly related to unit
size, the presence of scrubbers, and regional wages of construction
labor. Supercritical units were found to be considerably more costly
than comparable subcritical units. He also found average capital costs
varied depending on which A-E or utility built the unit. Perl argued
that this indicated the more experience the A-E had designing and
building generating units, the lower the capital cost of a unit.

Perl concluded that economies of scale are very limited for
subcritical coal-fired units. The higher average construction costs of
smaller units are offset by their higher equivalent availability. Thus,
the cost of electricity from subcritical units is basically constant
when unit size is beyond 200 megawatts. In contrast, Perl finds that
supercritical units are characterized by economies of scale. Average
capital costs fall with increases in size while availability remains
roughly constant. As a result, Perl concluded that subcritical units
are cheaper to build and operate than supercritical units when unit size
is less than 800 MW, and that the reverse is true when unit size

increases beyond 800 MW.

34

We believe there are a number of problems with Perl’s study. One
problem relates to the use of A-E dummy variables and Perl’s conclusion,
when these dummies prove to be statistically significant, that A-E
experience is a significant determinant of the capital cost of a unit.
A variable measuring experience must be included in the construction
cost equation to separate the effects of experience and any attributes
of particular A-Es. Characteristics of particular A-Es, such as design
philosophy, choice of vendors for major components, and the quality of
the units, may be correlated with experience. So both experience and
A-E specific characteristics must be included to avoid biased parameter
estimates.

Another problem relates to Perl’s failure to include availability
in the capital cost equation even though he recognizes that unit size
and equivalent availability are inversely related. This will cause the
parameter estimates to be biased and makes it difficult to determine
whether large units cost less to build due to economies of scale or
poorer quality.

Houldsworth (1985) examined the relative importance of capacity
utilization and plant size in determining the average cost of generating
electricity. He adopted the same quasi-engineering approach used by
Stewart,‘33 except for how the level of utilization is included in the
model. Recall that Stewart defined the expected cumulative output of a
generating plant as:

0 - 8760bk
Where,

8760 - number of hours in a year,

35

b - expected plant factor,

k - capacity of the plant measured in megawatts.
It is also important to recall that Stewart’s sample consisted of a
mixture of baseload and non-baseload plants which entered commercial
operation during the period 1970-71. He used the plant factor observed
in 1972 for each plant as a proxy for the expected plant factor.

Houldsworth recognized the desired level of utilization is a
significant determinant of generating plant construction costs. He also
recognized that the ex post plant factor will probably understate the
level of desired utilization. Plant size is positively correlated with
desired utilization intensity, while equivalent availability is
negatively correlated with unit size. Houldsworth also recognized that
mixing baseload and non-baseload plants in the same sample is
inappropriate since they have different levels of desired utilization.

Houldsworth avoided these mistakes by restricting his sample to 32
coal-fired baseload plants which began commercial operation in the
period 1972-1978. He also defined the level of utilization by the
expected availability, a (k), since each unit was assumed to serve
baseload demand. This meant that the expected cumulative output
relationship became:

0 - 8760a(k)K

Houldsworth combined the estimated plant cost function with the
plant size, equivalent availability, and factor prices to compute the
cost minimizing heat rate and the average total cost per kwh for each
plant. The simulations conducted by Houldsworth indicated average total

cost per kwh reached a minimum between 150 and 250 MW, and increased

36
moderately for larger sized units. Thus, Houldsworth concluded that
declining availability offset the benefits of any reductions in heat
rate and/or average construction cost per kw associated with units
larger than 250 megawatts.

Our primary concern lies with Houldsworth’s use of availability
data published regularly by the North American Electric Reliability
Council (NERC). These reports provide availability data that are
cumulative over a ten-year period and are averaged over many types of
units. The NERC data mixes units in the same size range even though
they have considerable differences that significantly affect their
reliability. Units within the same NERC size categories vary with
respect to fuel-type (gas, oil, or coal), desired level of utilization
(baseload or non-baseload), and in steam conditions (high or low steam
pressure and termperature). ”The reliability data given are, therefore,
average figures that reflect ’average’ hypothetical units.“”

Joskow and Rose (1985) are interested in determining what impact
unit size, differences in technology, tightened environmental
restrictions, and the experience of utilities and A-Es has on the costs
of building coal-fired generating units. They specify the construction

cost model:

LAC - :9?” AT, + b,LSIZE + bzRWAGE + baFIRST +

b,SCRUBBER + b5COOLTWR + DBUNCONV + b7EXPERAE +
b,EXPERU + b,EXPERI + S + U.
Where,

LAC - natural log of real cost per KW of a unit, net of AFUDC;

37

'L - one if the unit entered commercial operation in year t, and
zero otherwise;

LSIZE - natural log of unit size in megawatts;

RWAGE - regional average union wage for construction workers in
1976;

FIRST - one if the unit is the first unit on the plant site, zero
otherwise:

SCRUBBER - one if unit was built with a scrubber, zero otherwise;

COOLTWR - one if the unit was built with a cooling tower, zero
otherwise;

UNCONV - one if the unit is not of full-indoor construction, zero
otherwise;

EXPERAE - the cumulative number of "like” units designed by the A-E
being observed that entered commercial operation between 1950 and
year t;

EXPERU - utility experience since 1950;

EXPERI - total industry experience since 1950;

S - a seperate intercept term for each A-E.

The data set consisted of 411 coal-fired generating units which
entered commercial operation between 1960 and 1980. The units were
divided into four turbine throttle pressure groups: 1800, 2000, 2400,
and 3500 PSI. Subcritical units have steam pressures less than 3206
PSI, while supercritical units have steam pressure in excess of 3206
PSI. The units ranged in size from 100 megawatts to 1300 megawatts.

Joskow and Rose treat this data set as a panel with individual
generating units as observations over time on a cross section of A-Es.
They believe there are A-E specific design characteristics common to
units designed by a particular firm so they use fixed effects estimation

to control for these effects. Fixed effects means estimating a separate

intercept for each A-E.

38

It should be noted that Joskow and Rose specify a Cobb-Douglas
relationship between cost and unit size. This forces the cost function
to have a constant elasticity of unit cost with respect to size.
Additional flexibility is added by allowing the size coefficient to take
on a different value for each pressure group. The intercept is also
allowed to vary across pressure groups.

Joskow and Rose found significant differences between the cost
characteristics of subcritical and supercritical units. Supercritical
units are more costly to build than subcritical units except for large
unit sizes. Only when unit size increases beyond 600 megawatts do
supercritical units become cheaper to build than subcritical units. As
a result, they conclude there "is no simple static tradeoff between unit
size and construction cost: full exploitation of economies of scale in
construction costs can only be achieved by moving from one technology to
another. It would be wrong to think of static economies of scale
independently of choice of technology.”

They also found scrubbers and cooling towers added 15% and 6%,
respectively, to the construction cost of coal-fired units. Experience
was found to be numerically and statistically significant for
supercritical units only. Average construction cost for a supercritical
unit fell approximately 15% when the architect-engineer’s experience
with that type of unit increased from zero to the average level of
experience in the sample.

We believe there are a number of problems with the Joskow and Rose
study. One is the likelihood that baseload and non-baseload units, and

their different levels of desired utilization, were mixed in the sample.

39
Again, the level of desired utilization, independent of unit size, is an
important determinant of the cost of building a generating unit.
Failure to control for desired utilization and mixing baseload and
non-baseload units in the sample will bias the parameter estimates since
the level of desired utilization generally increases with unit size.

Joskow and Rose also note large units generally have lower
equivalent availabilities than small units and that "especially poor
performance is exhibited by the larger supercritical units."36 However,
they fail to treat reliability as an attribute of generating units and
thus exclude it from the construction cost function. As a result, it is
difficult to determine whether large units cost less because of
economies of scale or because of poorer quality.

Schmalensee and Joskow (1986) were concerned with how the
construction cost of a coal-fired generating unit varied with the
quality of the facility. The two indices of quality were the units’
heat rate and equivalent availability. They used a two-stage estimation
process. The first stage was concerned with obtaining estimates of
unit-specific quality attributes from data on actual unit performance
and operating characteristics that affect performance over time. A
fixed effects model was used to obtain estimates of each of the two
quality attributes which are supposed to enter the second stage of the
process, estimation of a construction cost function. The sample used by
Schmalensee and Joskow consisted of observations on 71 subcritical
coal-fired units which entered commercial operation between 1960 and

1969. The units ranged in size from 218 MW to 709 MW. The sample also

40
contained operating performance data, heat rate and equivalent
availability, for these units, for the years 1969 through 1977.

The first stage consisted of two performance equations of the
following form:

REL - 0:, + WY, + V,

EFF - 05, + WY, + V2
Where

REL - -ln(1 - equivalent availability),

EFF - -ln[(gross heat rate - 6000)/6000].

The W’s are matrices of exogenous variables that should affect
intra-unit variations in performance over time. The variables included
in W are:

CAPU - deviation in output factor [-100X generation/(capacity x

hours in service)] from the sample mean for all units, a measure of

relative capacity utilization,

AGE - unit age (calendar year - year unit entered commercial
operation) minus three,

BTU (MOIST, ASH, SULPH) - deviation of BTU’s per pound (percentage

of moisture, ash, sulpher) of the coal burned from the sample mean

for the observed unit.
Age entered each equation quadratically to allow for the possibility
that performance improves when a unit first enters commercial operation,
and then decreases as the unit ages beyond some point. The D’s are
matrices of unit--specific dummy variables that take on a value of one
for a unit when the observations are associated with that unit and zero
otherwise. Thus, the coefficients a, and 52 are the estimates of unit

specific quality attributes used as explanatory variables in the

construction cost equation.

41

Schmalensee and Joskow found unit performance deteriorated as units
aged after a short break-in period. Reliability peaked after a year or
less of Operation (note that age is measured as actual age minus three)
and then fell, while fuel efficiency declined from the start of
operations. Also, the coal characteristics variables were never
statistically significant. Thus, intra-unit variation in coal
characteristics appeared to have little effect on unit performance.

Schamalensee and Joskow rejected the null hypothesis that unit
qualities are identical. The estimated unit-specific coefficients of
REL (the estimated elements of 5,) correspond to equivalent
availabilities ranging from .6 to .97 and a mean of .84. The estimated
unit-specific coefficients of EFF (the estimated elements of :2)
correspond to heat rates ranging from 7,700 to 10,800 BTU/KWH and a mean
of 9,000.

The second stage consisted of estimating a construction cost
equation of the form:

AVCOST - f(SIZE, WAGE, BTU, TIME, EFF, REL).
Where,

AVCOST - natural log of unit capital cost in 1965 dollars per KW of
capacity,

SIZE - natural log of nameplate capacity in megawatts,

BTU - natural log of unit-specific mean of BTU’s per pound of coal
burned,

WAGE - natural log of regional construction wage in 1965,

EFF - design thermal efficiency of the unit (the estimated
unit-specific coefficients, :2, as defined in the first stage),

REL - design reliability of the unit (the estimated unit-specific
coefficients, 5,, as defined in the first stage).

42

Schmalensee and Joskow estimated two specifications of the
construction cost equation. The first specification was linear in the
variables, except time which was entered as a quadratic. The second
specification allowed the effects of reliability and fuel efficiency to
interact with the quality of the coal burned variable (BTU).

Each specification was estimated three different ways. One
ordinary least squares (OLS) estimate was obtained by using
unit-specific averages of observed heat rates and equivalent
availabilities as the quality variables rather than the values
(coefficients) estimated in the first stage. The second set of OLS
estimates was provided by using the values (coefficients) estimated in
the first stage. The third set of estimates was provided by an adjusted
least squares technique developed by Schmalensee and Joskow.

Schmalensee and Joskow found the quality attributes, EFF and REL,
were statistically insignificant, and frequently had implausible signs
and magnitudes. All six equations implied that increasing fuel
efficiency reduced construction costs per unit of capacity. The
adjusted least square estimates indicated REL had a positive impact on
costs, but that it was never close to statistical significance. In the
linear and interactive specifications estimated using OLS, REL had a
negative insignificant coefficient.

Schmalensee and Joskow found the size coefficients were
statistically significant in the OLS estimates, but statistically
insignificant in the adjusted least squares estimates. The magnitude of

the estimated size coefficients did not differ very much and implied

43
that doubling unit capacity would reduce average construction cost per
unit of capacity by between 10 and 12 percent.

Schmalensee and Joskow are the first to include quality variables
for both fuel efficiency and equivalent availability, but there are
still a number of problems with their study. One problem relates to
their inability to include a size variable in the first stage
regressions, since it would be perfectly collinear with the
unit-specific dummy variables.”’ Other things equal, equivalent
availability falls as unit size increases. Unit size is also negatively
correlated with construction cost per unit of capacity. Thus, use in
the construction cost equation of an equivalent availability variable
that fails to control for unit size may cause the parameter estimates to
be biased. They also recognize there may be architect-engineer (A-E)
specific variations in ex post performance given the level of
construction costs and design performance levels, but cannot capture
these effects explicitly due to a conflict between sample size and the
number of explanatory variablesf” Again, these effects will be
reflected in the fixed-effects estimates of the unit-specific quality
variables of the first stage.

M

A review of the literature reveals that unit size, technological
vintage, and fuel type have been viewed as the primary determinants of
the construction cost per KW of capacity of a generating unit. A
general conclusion of these studies is that there are economies of scale

with respect to construction costs.

44

But this review also shows few previous studies have properly
accounted for the desired utilization intensity of a generating unit. A
majority of the studies mix baseload and non-baseload units in the
sample and also fail to include the mode of operation (baseload or
non-baseload) as an explanatory variable, so the estimated size
coefficients are likely to be biased. Given that unit size and desired
utilization intensity are positively correlated, it’s likely the
estimated size coefficients are picking up the affect of moving from
non-baseload to baseload operation, and not just the impact of unit
size.

Also, the vast majority of these studies fail to include
reliability as an attribute or characteristic of generating units. They
ignore the ex ante design process in which reliability is a key
parameter. But the key issue is the intensity with which a unit is
used. A poor level of reliability, especially for baseload units, means
the capital-related costs of a unit are spread over a lower level of
output than if a unit is used more intensively. Thus, these studies
fail to recognize that an inexpensive unit to build may not be so
inexpensive to operate.

The studies by Perl, Houldsworth, and Schmalensee and Joskow (1986)
are the only studies that include in their analysis the effects of unit
reliability. Unfortunantly, each of these studies was flawed. Both
Perl and Houldsworth fail to include availability in the capital cost
equation even though they recognize that unit size and equivalent

availability are inversely related. This causes their parameter

45
estimates to be biased and makes it difficult to determine whether large
units cost less to build due to economies of scale or poorer quality.

Schmalensee and Joskow (1986) were the only researchers to include
equivalent availability as an explanatory variable in the construction
cost equation. However, they used an estimate of equivalent
availability that did not control for unit size. This is a significant
problem since unit size and equivalent availability are, in general,
negatively correlated. Thus, use of their equivalent availability
variable in the construction cost equation may bias the estimated
coefficient since it may be picking up the negative size effect.

This review indicates a construction cost model must recognize the
ex ante design and construction process in which reliability is a key
parameter. Not only must the construction cost function include
reliability as one of the attributes of a generating unit, but the model
should also recognize that unit reliability is a function of numerous
factors. The development of such a model is the subject of the next

chapter.

CHAPTER 3

DEVELOPNENT OF A GENERATING UNIT CONSTRUCTION
COST AND RELIABILITY NODEL

mm

In this chapter we develop a model of the steam electric generation
process that recognizes the multidimensional nature of capital embodied
in baseload coal-fired generating units. The first section, the
Heterogenous Nature of Capital, emphasizes how a utility’s choice of
unit characteristics at the ex ante or design stage determines the ex
post production relationships. Thus, the fuel-output relationship is
fixed once the unit is built and it is fairly insensitive to the rate of
generation. In the second section, the Model, we develop a model that
highlights the capital-intensive nature of baseload coal-fired
generation and the importance of intensive utilization if a unit’s
average total generation costs are to be minimized. In the third
section, Specification of the Construction Cost and Reliability Model,
we develop a simultaneous equation model of unit construction cost per
KW of capacity where average construction cost and reliability are
endogenous variables. In the fourth section, Empirical Analysis, the
data employed in the study is discussed and the estimation results are

presented.

46

47
W

- o f

The cyclical nature of the demand for electricity has important
implications for the nature of capital in the electric generating
industry. The output of a generating unit can be thought of as having
at least two dimensions. Power, the first dimension, is the
instantaneous rate of output and is measured in kilowatts (KW). The
second dimension is energy, measured in kilowatt hours (KWH). Energy is
the product of a power level and the period of time (measured in hours)
over which the unit operates at that level. Thus, energy is the
cumulative level of output over a period of time.

The demand for electricity fluctuates in a cyclical pattern over a
day, week, or season. Generally, however, electric power is
non-storable, so the supply of electricity must equal the demand for
electricity at all times. Thus, a utility must install sufficient
capacity to satisfy the maximum demand expected over the cycle. Given
that the demand for electricity varies more quickly than generation
capacity, a portion of the utility’s generation facilities will not be
operated at maximum capacity at all times. As a result, generating
units with the same maximum capacity may have different levels of
cumulative output and units with the same levels of cumulative output
may have different maximum capacities. In this environment it is
reasonable to expect a utility that is building a new generating unit to

consider both dimensions of output when making choices at the gt_nntg or

48
blueprint stage. Thus, the utility must decide what portion Of the
total demand for electricity the unit will serve. This decision will

then affect the size and desired thermal efficiency Of the planned unit.

MW

A utility faces a wide range Of production possibilities at the
blueprint stage. Each blueprint represents a generating unit with
different engineering characteristics such as unit size, unit type
(baseload and non-baseload), reliability requirements, fuel type and
quality, steam pressure conditions (an ex ante measure Of thermal
efficiencyafi, pollution control techniques, unit life, expected capital
cost, expected Operation and maintenance costs, and numerous other
characteristics. But once the generating unit is built, the fuel-output
and labor-output relationships are fixed. Thus, the utility faces a set
Of gt_nntg production possibilities from which various engineering
attributes are selected which shape the 95.29;; production
rel ationship.40

Labor is heavily dependent on the design Of the unit and thus
exhibits very little response tO variations in output. The Operational
labor requirements are affected by the number and size Of units in
Operation at a site. The choice Of fuel also affects the number Of
Operational personnel needed, since it determines the fuel handling
requirements. In particular, coal requires more labor input than does
either Oil or natural gas. The number Of maintenance personnel also
depends on the number and size Of units and the schedule Of routine

maintenance, rather than the level of output. Routine maintenance is

49

scheduled on an annual basis and is geared to unit size, unit type, and
the presence Of pollution abatement technologies such as scrubbers.‘1

The relationship between the flow Of fuel and output also depends
on the design characteristics Of the generating unit. Thermal
efficiency increases (heat rate falls) with the temperature and pressure
Of the steam, the thermal efficiency Of the boiler, the efficiency Of
the turbine, and the size Of the boiler and the turbine. Once the unit
is built, marginal fuel use, or the incremental heat rate, is related to
the capacity utilization Of the unit. According to Bushe (1981, Chapter
4) there are a number Of alternatives for the form Of the incremental
heat rate function for a single generating unit (see Figure 3-1). He
notes that, it is likely that the form depends on the design Of the unit
and large baseload units may have forms like (d) or (e) Of Figure 3-1,
while peaker units may have forms like (c). Since we are concerned only
with baseload units in this study, we will assume that the relevant
forms are (d) or (e). This assumption means the incremental heat rate
is fairly insensitive tO changes in load within the normal Operating
range Of a baseload unit. As a result, the amount Of fuel required is
proportional tO the rate at which the unit produces electricity.“Z

Thus, the fuel-output and labor-output relationships are
conditional upon the design characteristics of the generating unit. Once
the unit is built there is very little Opportunity for substitution
among the inputs, so the technology for generating electricity is

putty-clay.

50

 

 

 

 

 

 

di
.91. —
do do
(a) L] (b) q
.91.
dq
(c) q
.eL EL
dq dq

 

/' _/

(d) (1 (c) q

 

 

 

 

Figure 3.1. Possible Forms for the Incremental Heat
Rate

df/dq = incremental heat rate
q = load

Source: Bushe (1981 ), Chapter 4

51

ener

The traditional neoclassical production model assumes that input
capital is homogeneous (i.e., one unit Of capital services is equal to
any other unit Of capital services). This assumption implies that a
unit Of capital can be represented by a scalar measure such as size or
the dollar value Of the capital equipment. We believe, however, that
the heterogeneous nature Of capital means the electric generation
process cannot be accurately summarized by the traditional neoclassical
production function. The fuel-output and labor-output relationships are
tOO dependent on the characteristics Of the capital equipment to be
ignored!“ Therefore, we developed a model that explicitly recognizes
the heterogeneous nature Of capital in the electric generating industny.

Our concern is with the investment decision Of the utility since
the characteristics Of the capital equipment restrict the variable
input-output relationships or, in other words, the associated short-run
production possibility sets. The utility must decide what portion Of
total demand the new unit will serve and the associated engineering
attributes Of the unit.

We assume that the utility wants to serve an exogenous baseload and
will choose the engineering characteristics Of a coal-fired unit to
minimize the expected total cost Of meeting the load. A key issue here
is the intensity with which a unit is used. The capital intensive
nature Of baseload coal-fired generation means that the average total

generation costs Of a unit can be significantly reduced if the unit is

52

used intensively. Baseload units are meant tO be Operated at maximum
capacity continuously so that the level Of utilization is dependent on
the expected availability Of the unit. Thus, the expected output Of the
unit is defined as:

0 - 8760*A(X)*K,
Where,

0 - expected output, measured in kilowatt hours,

A(X) - expected availability Of the unit,

X - unit attributes that affect availability,

K . size Of the unit measured in kilowatts.

h c d P O un tiO

As noted earlier, electric generating technology is putty-clay.
The thermal efficiency Of a unit is variable at the design stage and
dependent on the utility’s choice of steam pressure conditions. But
once a unit has been built, the fuel-output relationship is fixed.“
Thus, the total cost Of generating electricity is a function Of the
level Of output, the price Of fuel on a BTU basis, the thermal
efficiency Of the unit, construction costs per KW of capacity, and the
annual fixed charge rated” The fixed charge rate is the annual cost Of
money capital including depreciation, property taxes, and other expenses
proportional to capital costs. As a result, expected §X_DQ§L total
generation costs are:

SRTC(P,.,P,,,r,¢,K) - .8760 A(X,,)K0

'PF + rPK(Zo)KO’

53
Where,
a = the unit's heat rate (BTU's/KWH),
PF = price per BTU Of fuel,
m2)

Z = unit attributes which affect construction costs,

unit construction costs per KW Of capacity,
r = the annual fixed charge rate.

Ex Ante Costs and the Investment Decision

Given that the technology is putty-clay, the utility is faced with
the task Of choosing unit attributes that minimize the expected total
costs Of serving a baseload demand for electricity at the blueprint
stage. The primary unit attributes to be chosen are unit size, steam
pressure conditions (an ex ante measure Of thermal efficiency), and the
reliability Of the unit. The average construction cost of a coal-fired
unit is also assumed to vary with the choice Of these engineering
characteristics. Briefly, it is reasonable to expect a movement from
subcritical to supercritical steam pressure conditions to increase
average construction cost. Higher pressure conditions improve thermal
efficiency,46 but necessitate the use Of costly materials capable of
handling the extreme pressure. It is reasonable to expect that aPk/aK
will take on positive, negative, or zero values depending on the size of
the unit. We expect that aPk/aK will take on negative values at low
unit sizes and possibly become positive at large unit sizes. We also
expect that the average construction cost will increase with the
reliability Of the unit, aPk/aA>O. Finally, we expect average
construction cost tO increase at an increasing rate as availability

approaches the limit of 100 percent, asz/3A2>O.47

54

As noted earlier, a unit's fuel costs per KWH are relatively
insensitive to the rate of generation.48 But the capital intensive
nature of baseload coal-fired generating units means that intensive
utilization Of a unit spreads annual capital costs (which are fixed once
a unit is built) over a greater level Of output than if a unit is used
less intensively. Thus, the ex ante total costs per KWH generated by a
coal-fired baseload unit are in large part determined by the
availability, or reliability, of the unit:

TC(Pf,r,o,K) + a876OKA(X)Pf+rPk(Z)K0.

This implies that the total generating costs per KWH Of a
subcritical or supercritical unit can be approximated by minimizing the
unit's annual capital cost per KWH produced:

Capital Cost = (Total Capital Cost x Annual Fixed Charge Rate)
per KWH Q

Since 0 = 8760 A(X)K,

Capital Cost = Capital Cost per KW * AFCR
per KWH 8760 * A(X)

This indicates that the ratio Of average capital cost to unit
availability determines capital cost per KWH generated. In general,
both availability and capital cost decline with increased unit size for
both subcritical and supercritical technologies. Thus, minimum total
generation costs per KWH can be approximated for a given type Of unit
(subcritical or supercritical) by selecting the level Of reliability and
unit size such that the ratio Of average capital cost to availability is

the least.

55

SPECIFICATION OF THE CONSTRUCTION COST AN RELIABILITY MODEL

 

The Determinants Of Unit Construction Cost

The construction cost Of a coal-fired generating unit depends on
the cost Of inputs and a number Of unit-specific attributes. Attributes
commonly thought tO influence the construction cost per KW of capacity
include unit size, unit order, steam pressure conditions, reliability,
and cooling method.

Reliability, measured by equivalent availability, is not the result
Of a single design or construction decision. Rather, reliability
encompasses a large number of design and construction decisions ranging
from the number and sizing Of various types Of equipment to the quality
of the inputs used to manufacture or assemble the various components Of
the unit. Above all, redundancy of critical components is necessary to
maintain a high level of unit reliability. A unit can continue in
operation when a critical component fails, or needs regular maintenance,
only if the component has a backup.49 As a result, one would expect
that increasing the level Of reliability, everything else constant,
requires additional capital investment which means aPk/aA>0. We also
expect that increasing reliability beyond some point will cause costs to
increase at an increasing rate. Thus, we expect asz/aA2>0.

According to engineering literature, construction costs increase
less than proportionately with the size Of the unit. As unit size
increases, the amount of material and labor per KW of capacity
decreases. For example, a 600 MW unit uses only twice the piping and
steel necessary to build a 200 MW unit.50 But it would also seem
reasonable to expect average construction costs to increase as unit size

increases beyond some point. Extremely large unit size might require

56
the use Of additional structural reinforcement, special materials or
construction methods. Thus, we expect that aPk/ak will take on negative
values at low unit sizes and possibly become positive at very large unit
sizes.

The primary technological frontier with respect to the thermal
efficiency Of coal—fired units built since 1960 has been in steam
pressure conditions.51 These units fall intO two major technological
classes--subcritical units with steam pressures below 3206 PSI and
supercritical units with pressures greater than 3206 PSI. Subcritical
units fall into three pressure classes around 1800 PSI, 2000 PSI, and
2400 PSI. These units require nO technological changes as steam
pressure in increased. But the higher pressures dO require thicker
casings for components and materials that can handle the higher
pressures. Supercritical units represent an entire different technology
since there is nO real boiling process and steam is produced
continuously as the temperature Of the water increases. The need for
some equipment associated with a conventional boiling process is
removed, but the large increase in pressure necessitates considerable
expenditure on special materials capable of withstanding these
pressures. Thus, we will use the design steam pressure condition as an
ex ante measure of thermal efficiency, and we expect supercritical units
to be more costly to build than subcritical units.

The majority of individual generating units are part Of multiunit

sites, and the order in which the units are built has considerable

57
impact on their individual construction costs. The first unit at a
multiunit site is usually built with sufficient waste disposal
facilities, transportation facilities, fuel-handling facilities, and
other common facilities to support the Operation Of all other units
scheduled to be built on the site at a later date. Thus, we expect the
first unit at a site tO be more costly than follow-on units.

A couple Of previous studies have examined the effects Of utility
and architect-engineer (A-E) experience on the construction costs Of
generating units. Joskow and Rose (1985) found that there are important
learning or experience effects as the number Of units built by a given
A-E increases. However, they also found that the experience effects
were limited to the building Of supercritical units. In order to
account for the presence Of any experience effects, we include a
variable which measures each A-E’s cumulative experience since 1950 with
a specific technology. We broadly define all subcritical units as
falling into one technology group, and all supercritical units as
following into another technology group.

In addition to the variables mentioned above, we also include a
number Of dummy variables. First, it is possible that there are
regional cost differences which might arise for a number Of reasons.

But the primary reason is that the market for construction labor is
generally regional and wages can vary considerably across regions.
These variations may be due to differences in the degree Of unionization

and the tightness Of the regional labor market.

58

Second, we include a dummy variable to indicate whether the unit is
Of full-indoor design. In colder climates, boilers and
turbine-generators are usually fully enclosed in protective structures.
But these facilities are Often only partially enclosed or fully outdoors
in warmer parts Of the United States. Thus, one would expect units Of a
full-indoor design tO be more costly to build.

Third, we include a dummy variable for each A-E so as to reduce or
eliminate any potential omitted variable bias. There are likely tO be
unobserved A-E specific design characteristics common to units designed
by a particular firm. These unmeasured attributes may be correlated
with experience or unit reliability which means that failing the account
for them could cause the parameter estimates tO be biased. Also, the
inclusion Of A-E dummy variables allows us to assess any differences
between specific A-E firms.”

A translog specification of the construction cost function will be
estimated. The primary reason for using this specification is that it
is flexible enough to allow all size-reliability effects to occur. The
basic construction cost relationship is:

lnPK - 80 + B,ln(MW) + Bz(ln(MW))2 + B,(-ln(1-EA)) +

B,(-ln(1-EA))2 + 85(-ln(1-EA))ln(MW) +

B;*PRESSURE + B7(PRESSURE * ln(MW)) + B,(PRESSURE * (-ln(1-EA))) +

B,ln(l + EXPERAE) + B,,,LN(YEAR) +

7.3.x, + e,.

Where,

PK - the real dollar cost per KW Of a generating unit net Of
capitalized interest costs,

MW - the capacity Of the unit in megawatts (MW),

59

EA = a three year average Of a unit's Observed equivalent availability
(The Observations covered each unit's second through fourth year of
commercial Operation.),

PRESSURE = a dummy variable which indicates whether the unit is
supercritical,

YEAR = the year the unit entered commercial Operation minus 1900,
EXPERAE = the cumulative experience since 1950 Of the A-E with the
technology of the Observed unit (If a unit enters commercial Operation
in the year t the total experience Of the A-E is measured as the total
number of units Of the same technology designed by the A-E that entered
commercial operation before year t.),

x = dummy variable indicating regionality, the presence of cooling
towers, the first unit at a multiunit site, and the other variables
discussed above,

et = the error term.

The Determinants Of Unit Reliability

The engineering and economic literatures indicate that a number Of
design and operational characteristics have a systematic effect on unit
availability. Among these are unit size, steam pressure conditions,
unit age, vintage, mode of Operation, and the degree Of redundancy of
key components.

Some engineering and economic studies assume that availability is
independent Of unit size,53 but several studies that examine the size-
availability relationship conclude that size and reliability are
inversely related.54 One explanation for the inverse relationship is
that similar types of outages are generally longer for large units than
for smaller units due to the larger area and parts that have to be
repaired. Also, large unit size means more material and equipment which
simply creates more opportunities for things to break down. Thus, we

expect aRel/aK<O. TO explore more fully the relationship between

60
availability and unit size, we allow size tO enter with a quadratic
specification.

Previous studies have generally found that higher steam pressure
conditions have been associated with higher forced outage rates. In
particular, supercritical units have been found to be considerably less
reliable than similar sized subcritical units.55 Thus, we expect
supercritical units, everything else equal, to be less reliable than
subcritical units.

Availability also varies with the age Of the unit. The engineering
literature assumes that the availability of a unit will improve the
first two or three years Of commercial operation due to a process Of
debugging and learning by plant Operators. Unit availability is then
expected to remain fairly constant for a number Of years before starting
a process Of slow decay as the unit ages. Econometric studies find
evidence Of, at most, a one-year break-in period."’6 These studies also
find that unit availability peaks after a year or less Of Operation and
then steadily declines. We control for the effects Of aging by using
unit-specific averages Of equivalent availability Observed at the same
stage Of the unit life cycle. The Observed equivalent availability data
covered each unit’s second through fourth year Of commercial Operation.
This avoids or minimizes any possible bias due tO break-in periods while
also minimizing variations due to differences in utility-specific
maintenance behavior.

Another reliability factor is the mode Of Operation. One would
expect that load-following with a generating unit will reduce
availability relative tO that which would be Obtained with baseload

61
Operation. Load-following or cycling places a considerable amount Of
wear and tear on a unit and thus increases the likelihood that the unit
will break down.“' We control for the mode Of Operation by including
only units that are 300 MW or larger. New generating units with
capacities Of at least 300 MW are almost certainly baseload units.

As mentioned earlier, the engineering literature indicates that
unit reliability is a function Of numerous design and construction
factors. These factors range from a ’conservative’ design philosophy
(which includes a high level Of redundancy Of key components) to the
quality Of the materials used to make various components Of the unit and
a careful inspection process during the construction phase. We are
unable tO Observe these factors directly, so we are forced to use a
proxy. We believe that construction cost per KW is a good proxy since
all these factors can be expected to increase construction costs.

We also include the experience of the A-E with the given
technology. Designing and building a baseload generating unit is an
immensely complicated task, so it seems reasonable that A-E’s gO through
a learning process associated with the repetitive design and
construction Of technologically-similar generating units.

In addition to the variables mentioned above, we also include dummy
variables to control for any reliability factors that may be associated
with the different manufacturers Of key subcomponents, such as boilers
and turbine-generators, and any omitted A-E specific effects which may
bias the parameter estimates.

First, we include dummy variables tO indicate the manufacturers Of

the boilers and turbine-generators. It is reasonable to expect that

62

availability varies among the manufacturers Of these key subcomponents
due to differences in the design Of the equipment, the production
process, and the quality Of the inputs used in the production process.

We also include a dummy variable for each A-E (except one tO avoid
perfect multicollinearity) in an effort to reduce or eliminate any
potential omitted variable bias. Again, there are likely tO be
unobserved A-E specific characteristics which may be correlated with
explanatory variables, such as the choice Of boiler or turbine-generator
manufacturers. As a result, failure to control for these unobserved A-E
specific characteristics may bias the estimates Of the parameter vector.
This also allows us to assess any differences between specific A-E
firms.

The basic reliability relationship is:

-ln(l-EA) - A0 + A,ln(P,,) + A,ln(MW) +

A3(ln(PK)*ln(MW)) + AgPRESSURE + A5(PRESSURE*ln(MW)) +

A,ln(1 + EXPERAE) + A,ln(YEAR) +

mNX,+'e,
Where,

X - dummy variables indicating the first unit at a multiunit site,

manufacturers Of boilers and turbine generators, and the other

variables discussed above.

The reliability variable was specified so that the variable becomes

infinite as availability approaches the highest levels that are
theoretically possible.“

63
WW

As we have argued above, unit engineering attributes are determined
at the same time as unit construction cost. If the utility has some
conception Of the error term it faces in the construction cost function,
the utility might be expected to modify its choice Of unit attributes.
This means that the choice Of unit attributes may be correlated with the
error term. As a result, ordinary least squares estimates Of the unit
attribute coefficients in the cost function will be biased.

This estimation error will occur only if the utility has some
knowledge Of the error term it faces. However, it would seem
unreasonable to assume that the utility or A-E is totally unaware Of the
direction and size Of the error term. After all, every major utility or
A-E has considerable experience participating in the construction and/or
Operation Of different types Of generating units. Also, a utility or
A-E considering building a generating unit can draw on the experience of
other utilities that have recently built similar units.

The endogenous nature Of unit reliability in the cost function has
been hinted at by previous researchers. Komanoff (1976) notes that
higher initial construction cost due to a conservative design philosophy
and, thus, higher unit reliability may result in lower total generation
costs over time. Schmalensee and Joskow (1986) argue that it is only
"logical to assume that construction costs will vary with the ’quality’
Of the facility," and that one measure of quality is the reliability Of
the unit. Engineering case studies indicate that improved unit
reliability requires additional capital expenditures. High availability

requires that critical components have spares so that if one component

64
fails or service is required, the unit can continue in Operation. Thus,
high availability necessitates redundancy, and redundancy means higher
capital costs.“

Also, the reliability variable, EA, is the realized reliability
which consists Of the gx_nntg level Of reliability and a measurement
error. It is the et_nntg level Of reliability which determines cost Of
construction. TO ignore the fact that we can measure ex ante
reliability only with some error from actual performance data would
cause OLS estimation Of the cost function tO be biased.“0 In this
situation, an instrumental variables estimator, such as two-stage least
squares (ZSLS), is consistent.61

As a result, the construction cost and reliability equations are
treated as a simultaneous equations system:

lnP, - 80 + B,ln(MW) + B,(ln(MW))2 + B,(-ln(1-EA)) +

B,(-ln(1-EA))2 + 85(-ln(1-EA))ln(MW) +
B,*PRESSURE + B7(PRESSURE * ln(MW)) +
B,(PRESSURE * (-ln(1-EA))) + B,ln(1+EXPSUB) +
Bwln(1+EXPSUP) + B"Tn(YEAR) + BmFIRST +
BmTOWER + B“FULLIN + :BREGION,+

:BIAEDUM, + e,

-ln(1-EA) - A0 + A,ln(P,,) + A,ln(MW) +
A,(ln(P,,)*ln(MW)) + A,PRESSURE + A5(PRESSURE * ln(MW) +
A,ln(l + EXPSUB) + A,ln(l + EXPSUP) + A,ln(YEAR) +
AgFIRST + zA,BOILER MANU, + aA,TURBINE MANUI +

zAkAEDUM, + e,

65
Where,

FIRST - one if the unit is the first unit on the plant site, zero
otherwise,

TOWER - one if the unit was built with any type Of cooling tower,
FULLIN - one if the unit was a full indoor design, zero otherwise,
REGION - a regional dummy variable,

AEDUM - one if built by a particular architect-engineer. All
other variables are as previously defined.

Boiler and turbine manufacturer dummy variables are included in the
reliability equation but excluded from the cost equation. This was done
because a review Of the literature found that differences in component
design across manufacturers could affect reliability while having no
systematic affect on unit construction costs. The Tower and Fullin
variables are included in the cost equation but excluded from the
reliability equation. The presence Of a cooling tower and a fully
enclosed boiler increases the number Of structures tO be built which
means higher construction costs; however, a review Of the engineering
literature found no reason why the presence Of either design attribute
should affect unit reliability. Finally, regional dummy variables were
included in the cost equation and excluded from the reliability
equation. Construction cost might vary regionally because construction
labor is hired in regional markets, across which wages vary
considerably. There is little reason to believe that unit reliability
varies on a regional basis.

The constant term, unit order, regional, and time effects are
assumed tO be exogenous in the construction cost equation. The effects

Of unit size, steam pressure conditions, cooling method, full-indoor

66
design, choice Of architect-engineers, and experience are assumed tO be
exogenous, since there seems tO be no direct way that unit construction
cost can affect the choice Of these attributes and/or information
regarding possible instruments is unavailable.

The situation with respect tO the reliability equation is quite
similar to that Of the cost equation. The constant term, time, and unit
order effects are assumed to be exogenous in the reliability equation,
while the effects Of size, pressure, experience, choice Of A-E, and
choice Of boiler and turbine manufacturers are assumed tO be exogenous
because there is no Obvious way that reliability can affect a utility’s
choice Of these attributes and/or information regarding possible

instruments is not available.

Estimation Of the Simultanggug-Eguatign; finds]

The construction cost and reliability equations are linear in the

 

parameters but nonlinear in the endogenous variables. The variables
(-ln(1-EA))2, (-ln(1-EA))*ln(MW), Pressure*(-ln(1-EA)), and
ln(PK)*ln(MW) are all nonlinear endogenous variables, since (-ln(1-EA)
and ln(PK) are endogenous variables. Drawing on the procedure developed
by Kelejian (1971) and used by Farber (1981) and Martin (1979), we
derive first-stage estimates Of the endogenous variables by including as
instrumental variables the squares Of all non-dummy exogenous variables.
Thus, the reduced form equations are approximated by a second-order
polynomial Of the exogenous variables. The predicted values from the
first stage were then used tO estimate the structural equations. This

procedure leads tO consistent estimates Of the parameter vectors using

67
two-stage least squares. Both structural equations are identified using

the criteria derived by Kelejian for nonlinear models.“2

W

111L931;

We estimated the above equations on a sample Of 84 coal-fired units
that entered commercial Operation between 1960 and 1974. The sample
consisted of 40 supercritical units and 44 subcritical units. The
supercritical units range in size from 359 MW to 1,300 MW and the
subcritical units range from 310 MW to 745 MW.

Total construction costs per generating unit were derived from the
U.S. Department Of Energy’s Steam-Electric Plant Construction Cost and
Annual Production Expenses (DOE). However, these cost data are reported
in nominal dollars and include interest charges capitalized during
construction. Therefore, we use a procedure implemented by Joskow and
Rose (1985) and Zimmerman (1982) to deflate for input price changes and
to remove the capitalized interest charges.

The deflation process is complicated by a number Of problems.
First, not only do the reported construction costs include interest
charges, but they reflect the summation Of nominal dollars spent over a
number Of years. Second, construction times and construction cash flow
profiles are not reported for individual units. SO we are forced tO use
a typical cash flow profile for units built in the early 1970’s. This

standard cash flow profile is combined with a price index and historical

68
interest rates to derive real construction costs net Of interest
charges.”

We used the Handy Whitman Index Of Public Utility Construction
Costs to deflate the nominal construction costs to constant 1973
dollars.“ The index is a proprietary seven-region index Of
steam-generating unit construction costs.

It should be noted that time is not a measure Of the impact Of
technological change on unit construction cost in this study. As noted
in Chapter 2, the primary goal Of technological innovation has been to
improve the thermal efficiency Of the generation process. The desire
for increased thermal efficiency has led to continued efforts to
increase steam pressure and temperature conditions. However, the
movement tO higher pressure (more technologically advanced) units
occurred only gradually: thus, a number of technologies are in use at
any point in time. This means that the use Of vintages or time to
define periods of technologically homogeneous capital is likely to be
inadequate.65

Furthermore, the primary technological frontier with respect tO
baseload coal-fired units built since 1960 has been in the steam
pressure conditions. These units fall into two major technological
classes--subcritical units with steam pressures below 3206 PSI and
supercritical units with pressures greater that 3206 PSI. Supercritical
technology is the primary technological innovation with respect to
thermal efficiency since 1960.” Thus, we include a supercritical
technology dummy variable to measure the impact Of technological change

on unit construction cost.

69

The equivalent availability variable was derived from data
collected by the National Electric Reliability Council (NERC).“' The
NERC data contained annual Observations Of equivalent availability for a
large number Of large (300 MW through 1300 MW) coal-fired units covering
the period 1965 tO 1977. Unfortunately, the units were Observed at
different points in the unit life cycle. As a result, some units were
Observed from their first year Of commercial operation: others were not
Observed until they had been in Operation for a number Of years. In
order tO Observe the units at the same point in the life cycle, we
included only those units for which we had Observations covering their
second through fourth year Of commercial Operation. The three annual
unit-specific equivalent availabilities were then averaged to derive our
reliability variable.”

Cooling tower information was Obtained from the Department Of
Energy’s Generating Unit Reference File (GURF). Architect-engineer
information for coal-fired units built since 1950 was Obtained from
annual survey’s in Power. Information on boiler and turbine
manufacturers was also Obtained from Power. All other data was Obtained

from DOE.

W

In order tO show the importance Of simultaneous equation bias
and/or measurement error we use both OLS and ZSLS tO estimate the cost
and reliability equations. Four variations Of the construction cost
function are estimated using both regression techniques. One form Of

the cost function includes ln(MW), (ln(MW))z, -ln(1-EA), (-ln(1-EA))2,

70
and an interaction term, ln(MW)*(-ln(1-EA)). The second form Of the
cost equation adds the interaction variable Pressure*ln(MW). The third
form eliminates the Pressure*ln(MW) variable and adds the
Pressure*(-ln(l-EA)) variable. The fourth cost function form estimated
includes both the Pressure*(-ln(l-EA)) and Pressure*ln(MW) variables.

Following Joskow and Rose (1985), we initially estimate the cost
function using ordinary least squares while excluding the reliability
variables and including dummy variables for all A-E’s except one (Stone
& Webster). This leaves us with only two specifications Of the cost
function. One specification includes the size variables ln(MW) and
(ln(MW))z, while the other specification adds the interaction term
Pressure*ln(MW).“’ Thus, the first specification forces the scale terms
for subcritical and supercritical units to be the same while the
intercepts for the twO technologies are allowed tO be different. The
introduction of the interaction term, Pressure*ln(MW), in the second
specification allows both the intercepts and scale effects to differ
across the two technologies.

The first-unit variable has a positive coefficient that is
significant at the 1% level in both estimates.7o This is consistent
with the fact that utilities have strong incentives to assign as much Of
the common costs Of multiunit sites as possible tO the first unit at the
site. Also, the time trend variable, ln(Year), has a positive
coefficient that is significant at the 1% level in both estimates. This
would seem tO indicate that units which entered commercial Operation at
later dates cost more to build than similar units which entered

commercial Operation earlier.

71

The A-E effects are jointly significant at the 5% level [F(15,54) -
2.24] and 10% level [F(15,53) - 1.91], respectively. This indicates
there are significant unobserved architect-engineer specific attributes.
The unit size variables are jointly significant at the 5% level in both
estimates [(F(2,54) - 4.48 and F(3,53) - 3.16, respectively]. Finally,
the null hypothesis that the intercept and scale terms for subcritical
and supercritical units are the same (F(2,53) - .35) cannot be rejected
(see Table A-5).

Next, we use OLS tO estimate four specifications Of the cost
equation when the reliability variables are added as mentioned above.
The estimated parameter values are presented in Tables 3-1 through 3-4.

An examination Of the cost function parameter estimates reveals
several things Of interest. The first-unit variable has a positive
coefficient and is significant at the 1% level in all four estimates.
The time trend variable, ln(Year), always has a positive coefficient and
is statistically significant at the 1% level in all four estimates. It
is interesting to note that the reliability variables are never jointly
significant. This occurs despite the -ln(1-EA) term being individually
significant in three Of the four estimates and the interaction term,
ln(MW) * (-ln(1-EA)), being significant in twO Of the four estimates.
Finally, we are unable to reject the null hypothesis that the intercept,
scale, and reliability terms for subcritical and supercritical units are
the same for the estimates reported in Tables 3-2 through 3-4
[F(2,50) - .27, F(2,50) - .2, and F(3,49) - .42, respectively].

72

Table 3-1. Ordinary Least Squares Estimate of the
Construction Cost Function

 

 

VARIABLE GREEEIGIEBDZ SIANQABD_EBBQB
CONSTANT -1.835 12.808
PRESSURE DUMMY - .148 .325
LN(MW) -1.205 3.215
(LN(MH))2 .001 .230
-LN(1-EOUIV. AVAIL.) -4.591 2.555-
(-LN(1-EOUIV. AYAIL.))2 .237 .142
LN(MW)(-LN(1-EOU1V. AVAIL.)) .614 .353-
FIRST-UNIT DUMMY .285 .055*
LN(YEAR) 3.509 .870*
COOLING TOWER DUMMY .084 .064
FULL-INDOOR DESIGN DUMMY .030 .085
A-E SUBCRITICAL UNIT EXPERIENCE - .038 .059
A-E SUPERCRITICAL UNIT EXPERIENCE - .076 .098
MIDDLE ATLANTIC DUMMY .004 .114
WEST NORTH CENTRAL DUMMY - .020 .149
EAST NORTH CENTRAL DUMMY .293 .259
ROCKY MOUNTAIN DUMMY - .080 .166
SOUTH ATLANTIC DUMMY - .168 .096-
SOUTHERN SERVICES CO. DUMMY - .286 .122+
TVA CO. DUMMY .074 .160
DUKE CO. DUMMY - .176 .162
AEP c0. DUMMY .202 .215
STEARNS a ROGER CO. DUMMY - 245 .244
BECHTEL CO. DUMMY - 109 .136
EBASCO CO. DUMMY - 195 .104-
SARGENT a LUNOY C0. DUMMY - 148 .103
BROWN 1 ROOT co. DUMMY - 210 .302
FLUOR CO. DUMMY - 569 .185*
BLACK & VETCH CO. DUMMY - 337 .170-
GILBERT CO. DUMMY - 168 .132
GC co. DUMMY - 039 .153
UNITED c0. DUMMY - 318 .262
COMMON co. DUMMY - 212 .147

 

* - Significant at 1 percent
+ - Significant at 5 percent
- - Significant at 10 percent
i? - .739

Adjusted R2 - .575

F(32, 52) - 4.505

 

73

Table 3-2. Ordinary Least Squares Estimate Of the
Construction Cost Function

 

 

ENABLE WW
CONSTANT - .254 13.186
PRESSURE DUMMY - .875 1.816
LN(MW) -1.962 3.496
(LN(MH))2 .073 .263
PRESSURE*LN(MW) - .164 .286
-LN(1-EOUIV. AVAIL.) -4.217 2.653
(-LN(1-EOUIV. AVAIL.))2 .232 .143
LN(MW)(-LN(l-EOUIV. AVAIL.)) .554 .370
FIRST-UNIT DUMMY .284 .056*
LN(YEAR) 3.587 .887*
COOLING TOWER DUMMY .081 .065
FULL-INDOOR DESIGN DUMMY .029 .086
A-E SUBCRITICAL UNIT EXPERIENCE - .019 .068
A-E SUPERCRITICAL UNIT EXPERIENCE - .072 .099
MIDDLE ATLANTIC DUMMY .010 .115
WEST NORTH CENTRAL DUMMY - .021 .150
EAST NORTH CENTRAL DUMMY .277 .262
ROCKY MOUNTAIN DUMMY - .061 .171
SOUTH ATLANTIC DUMMY - .165 .097-
SOUTHERN SERVICES CO. DUMMY - .280 .123+
TVA CO. DUMMY .070 .161
DUKE C0. DUMMY - .159 .166
AEP CO. DUMMY .210 .217
STEARNS a ROGER CO. DUMMY - 242 .246
BECHTEL CO. DUMMY - 123 .139
EBASCO CO. DUMMY - 189 .105-
SARGENT a LUNOY CO. DUMMY - 150 .104
BROWN 1 ROOT CO. DUMMY - 211 .305
FLUOR CO. DUMMY - 551 .189*
BLACK & VETCH CO. DUMMY - 318 .174-
GILBERT CO. DUMMY - 175 .134
GC CO. DUMMY - 031 .155
COMMON CO. DUMMY - 208 148
UNITED c0. DUMMY - 309 264

 

* - Significant at 1 percent
+ - Significant at 5 percent
- - Significant at 10 percent
i? - .74

Adjusted R2 - .569

F(33, 51) - 4.321

 

74

Table 3-3. Ordinary Least Squares Estimate Of the
Construction Cost Function

 

 

MOLE W WANDA BRO
CONSTANT .247 13.867
PRESSURE DUMMY - .054 .399
LN(MW) -1.662 3.426
mm»2 .026 .240
-LN(1-EOUIV. AVAIL.)*PRESSURE - .096 .234
-LN(l-EOUIV. AVAIL.) -5.172 2.937-
(-LN(1-EOUIV. AVAIL.))2 .234 .143
LN(MW)(-LN(1-EQUIV. AVAIL.)) .711 .427
FIRST-UNIT DUMMY .285 .056*
LN(YEAR) 3.464 .884*
COOLING TOWER DUMMY .081 .065
FULL-INDOOR DESIGN DUMMY .028 .086
A-E SUBCRITICAL UNIT EXPERIENCE - 041 .060
A-E SUPERCRITICAL UNIT EXPERIENCE - 086 .102
MIDDLE ATLANTIC DUMMY - 008 118
WEST NORTH CENTRAL DUMMY - .033 .153
EAST NORTH CENTRAL DUMMY .293 .261
ROCKY MOUNTAIN DUMMY - .098 .173
SOUTH ATLANTIC DUMMY - .181 .102-
SOUTHERN SERVICES CO. DUMMY - .295 .124*
TVA c0. DUMMY .065 .163
DUKE CO. DUMMY - .168 .164
AEP CO. DUMMY .190 .219
STEARNS a ROGER CO. DUMMY - 259 .248
BECHTEL c0. DUMMY - 108 .137
EBASCO CO. DUMMY - 198 .105-
SARGENT & LUNOY CO. DUMMY - 155 .106
BROWN 1 ROOT CO. DUMMY - 238 .312
FLUOR C0. DUMMY - 595 .197*
BLACK 1 VETCH c0. DUMMY - 363 .182-
GILBERT CO. DUMMY - 174 .134
CC C0. DUMMY - 054 .158
COMMON C0. DUMMY - 242 165
UNITED C0. DUMMY - 340 269

 

* - Significant at 1 percent
+ - Significant at 5 percent
- - Significant at 10 percent
1? - .74

Adjusted R2 - .568

F(33, 51) - 4.303

 

75

Table 3-4. Ordinary Least Squares Estimate Of the
Construction Cost Function

 

 

IABIAELE CQEEEIQENI §__BQ_L__TANDA RROR
CONSTANT 6.237 15.304
PRESSURE DUMMY 2.047 2.292
LN(MW) -3.765 4.107
(LNIMN))2 .200 .304
-LN(1-EOUIV. AVAIL.) -5.277 2.943-
(-LN(1-EOUIV. AVAIL.))2 .223 .144
LN(MW)(-LN(1-EOUIV. AVAIL.)) .736 .429-
-LN(1-EOUIV. AVAIL.)*PRESSURE - .233 .276
PRESSURE*LN(MW) - .315 .338
FIRST-UNIT DUMMY .282 .056*
LN(YEAR) 3.551 .890*
COOLING TONER DUMMY .071 .066
FULL-INDOOR DESIGN DUMMY .025 .086
A-E SUBCRITICAL UNIT EXPERIENCE - .009 .069
A-E SUPERCRITICAL UNIT EXPERIENCE - .093 .103
MIDDLE ATLANTIC DUMMY - .016 .119
NEST NORTH CENTRAL DUMMY - .053 .155
EAST NORTH CENTRAL DUMMY .261 .263
ROCKY MOUNTAIN DUMMY - .087 .174
SOUTH ATLANTIC DUMMY - .196 .104-
SOUTHERN SERVICES C0. DUMMY - .295 .125+
TVA CO. DUMMY .043 .165
DUKE CO. DUMMY - .123 .172
AEP C0. DUMMY .187 .220
STEARNS a ROGER C0. DUMMY - .272 .249
BECHTEL C0. DUMMY - .134 .140
EBASCO CO. DUMMY - .190 .105-
SARGENT & LUNDY CO. DUMMY - .169 .107
BROWN G ROOT CO. DUMMY - .279 .316
FLUOR CO. DUMMY - .597 .197*
BLACK & VETCH C0. DUMMY - .361 .182-
GILBERT CO. DUMMY - .195 .136
CC C0. DUMMY - .057 .158
COMMON CO. DUMMY - .275 .169
UNITED CO. DUMMY - .354 .270

 

* - Significant at 1 percent
+ - Significant at 5 percent
- - Significant at 10 percent
R2 - .744

Adjusted R2 - .567

F(34, 50) - 4.191

 

76

The OLS estimate Of the parameter values of the reliability
equation are reported in Table 3-5." The architect-engineer specific
effects are jointly significant at the 1% level (F(15,55)-2.83). The
unit size variable, ln(MW), has a negative and statistically significant
coefficient. There is also a negative and significant coefficient Of
the unit construction cost variable. The interaction between unit size
and unit construction cost has a positive and significant coefficient.
Finally, it is curious that the first-unit dummy has a positive and
significant coefficient.

The results reported in Tables 3-1 through 3-5 may be biased due to
the simultaneous nature of our model and/0r measurement error associated

with the use Of the realized level Of reliability instead Of the ex ante

 

level Of reliability. Two-stage least squares estimates are consistent
under these circumstances and appear in Tables 3-6 through 3-10.

A review Of the ZSLS estimates of the four specifications Of the
cost equation reveals a number Of differences when compared to the
corresponding OLS parameter estimates."2 The ZSLS estimates Of the
coefficient of -ln(1-EA) are always negative and reach some level Of
statistical significance in three Of four specifications. The
coefficient of (-ln(1-EA))2 is positive and significant at the 5% level
in all four ZSLS estimates Of the cost equation. These results imply
that the elasticity Of average construction cost with respect tO unit
reliability varies depending on the level Of reliability. The
interaction term, ln(MW)*(-ln(1-EA)), always has a positive coefficient
and is significant at the 5% level in two Of the four ZSLS estimates.
It is important to note that the ZSLS estimates Of the coefficients of
-ln(1-EA), (-ln(1-EA))2, and ln(MW)*(-ln(1-EA)) are generally

77

Table 3-5. Ordinary Least Squares Estimate Of the
Reliability Function

 

 

VARIABLE CQEEEIEIENI W
CONSTANT 35.278 13.249+
LN(PER KW CONSTRUCTION COSTS) - 7.14 2.782+
LN(MW) - 6.062 1.995*
LN(MW)*LN(PER KW CONST. COSTS) 1.071 .436+
LN(YEAR) 1.379 1.349
FIRST UNIT DUMMY .210 .093+
PRESSURE DUMMY - 1.180 1.743
PRESSURE*LN(MW) .150 .271
BOILER MANU. COMBUSTION ENG. .173 .118
TURBINE MANU. GENERAL ELEC. .376 .252
TURBINE MANU. WESTINGHOUSE .079 .250
BOILER MANU. BABCOCK & WILCOX .216 .142
A-E SUBCRITICAL UNIT EXP. .036 .087
A-E SUPERCRITICAL UNIT EXP. - .030 .136
SOUTHERN SERVICES CO. DUMMY - .114 .172
AEP CO. DUMMY .474 .291
DUKE C0. DUMMY .420 .182+
TVA CO. DUMMY .262 .288
STEARN & ROGER CO. DUMMY - 494 .293-
BECHTEL C0. DUMMY - 161 .144
EBASCO C0. DUMMY - 139 .136
SARGENT & LUNDY CO. DUMMY - 087 .151
BROWN 8 ROOT CO. DUMMY - 025 .421
FLUOR C0. DUMMY - 773 .268*
BLACK & VETCH CO. DUMMY - 203 .238
GILBERT C0. DUMMY - 060 .168
COMMON CO. DUMMY - 306 .204
UNITED CO. DUMMY - .546 .293-
GC CO. DUMMY .425 .221-

 

* - Significant at 1 percent

+ - Significant at 5 percent

62. Significant at 10 percent
- .8

Adjusted R2 - .697

F(28, 56) - 7.808

 

78

Table 3-6. Two-Stage Least Squares Estimate Of the

Construction

Cost Function

 

VARIABLE

CONSTANT

PRESSURE DUMMY

LN(MW)

(Lfllhflli’

-LN(l-EQU1V. AVAIL.)
(-LN(l-EOUIV. AVAIL.))2
LN(MN)(-LN(1-EOUIV. AVAIL.))
FIRST UNIT DUMMY

LN(YEAR)

COOLING TONER DUMMY
FULL-INDOOR DESIGN DUMMY
A-E SUBCRITICAL UNIT EXPERIENCE
A-E SUPERCRITICAL UNIT EXPERIENCE
MIDDLE ATLANTIC DUMMY

WEST NORTH CENTRAL DUMMY
EAST NORTH CENTRAL DUMMY
ROCKY MOUNTAIN DUMMY

SOUTH ATLANTIC DUMMY
SOUTHERN SERVICES CO. DUMMY
TVA C0. DUMMY

DUKE CO. DUMMY

AEP CO. DUMMY

STEARNS a ROGER CO. DUMMY
BECHTEL C0. DUMMY

EBASCO C0. DUMMY

SARGENT G LUNOY CO. DUMMY
BROWN 3 ROOT C0. DUMMY
FLUOR CO. DUMMY

BLACK G VETCH C0. DUMMY
GILBERT C0. DUMMY

GC CO. DUMMY

COMMON C0. DUMMY

UNITED C0. DUMMY

 

GOEEEIGIEMI W
29 877 19.462
- 124 .412
-8 283 4.646-

425 .323
-14 260 4.478*
712 .282+
1 953 .623*
288 .063*
2 784 1.197+
133 .081
013 .101
- 105 .120
- 071 .077
- 015 .139
040 .217
365 .311
- 036 .203
- 170 .115
- 296 .144+
179 .192
- 381 .240
- 211 .322
- 446 .314
- 168 .167
- 260 .137-
- 129 .125
- 375 .373
- 560 .252+
- 380 .201-
- 199 153
- 048 133
- 303 189
- 495 331

 

* - Significant at 1 Percent
+ - Significant at 5 Percent
- - Significant at 10 Percent
R2 - .6996

Adjusted R2 - .5112

F(32, 52) - 3.712

 

79

Table 3-7. Two-Stage Least Squares Estimate Of the
Construction Cost Function

 

 

IARLABLE BQEEEIBIEE WAN A RROR
CONSTANT 32.867 21.485
PRESSURE DUMMY - 2.121 3.320
LN(MW) - 7.978 5.018
(LN(MN))2 .364 .361
PRESSURE*LN(MW) .315 .519
-LN(1-EQU1V. AVAIL.) -16.005 5.606+
(-LN(1-EQUIV. AVAIL))2 .692 .3044
LN(MN)(-LN(1-E0UIV. AVAIL.)) 2.255 .833*
FIRST UNIT DUMMY .291 .068*
LN(YEAR) 2.186 1.620
COOLING TONER DUMMY .163 .099
FULL-INDOOR DESIGN DUMMY .014 .109
A-E SUBCRITICAL UNIT EXP. - .129 .135
A-E SUPERCRITICAL UNIT EXP. - .121 .117
MIDDLE ATLANTIC DUMMY - .040 .155
NEST NORTH CENTRAL DUMMY .098 .252
EAST NORTH CENTRAL DUMMY .407 .341
ROCKY MOUNTAIN DUMMY - .063 .223
SOUTH ATLANTIC DUMMY - .195 .131
SOUTHERN SERVICES C0. DUMMY - .333 .166-
TVA C0. DUMMY .173 .207
DUKE CO. DUMMY - .443 .277
AEP CO. DUMMY - .368 .433
STEARN a ROGER CO. DUMMY - .501 .349
BECHTEL CO. DUMMY - .184 .181
EBASCO CO. DUMMY - .319 .176-
SARGENT a LUNOY CO. DUMMY - .150 .139
BROWN 5 ROOT CO. DUMMY - .435 .412
FLUOR C0. DUMMY - .616 .285+
BLACK 3 VETCH CO. DUMMY - .455 .249-
GILBERT CO. DUMMY - .205 .165
G6 CO. DUMMY - .055 .202
COMMON CO. DUMMY - .341 .213
UNITED CO. DUMMY - .601 .397

 

 

* - Significant at 1 percent
+ - Significant at 5 percent
- - Significant at 10 percent
R2 - .6738

Adjusted R2 - .4584

F(33, 51) - 3.129

80

Table 3-8. Two-Stage Least Squares Estimate of the

Construction Cost Function

 

 

VARIABLE

CONSTANT

PRESSURE DUMMY

LN(MN)

(”i (W) )

-LN(1- -EOUIV. AVAIL. )*PRESSURE
-LN(1 EQUIV. AVAIL.)
(-LN(1 EQUIV. AVAIL))
LN(MW)( LN(1- EQUIV. AVAIL))
FIRST UNIT DUMMY

LN(YEAR)

COOLING TONER DUMMY
FULL-INDOOR DESIGN DUMMY
A-E SUBCRITICAL UNIT EXP.
A-E SUPERCRITICAL UNIT EXP.
MIDDLE ATLANTIC DUMMY

NEST NORTH CENTRAL DUMMY
EAST NORTH CENTRAL DUMMY
ROCKY MOUNTAIN DUMMY

SOUTH ATLANTIC DUMMY
SOUTHERN SERVICES CO. DUMMY
TVA CO. DUMMY

DUKE CO. DUMMY

AEP CO. DUMMY

STEARNS 1 ROGERS CO. DUMMY
BECHTEL CO. DUMMY

EBASCO C0. DUMMY

SARGENT a LUNOY CO. DUMMY
BRONN G ROOT C0. DUMMY
FLUOR CO. DUMMY

BLACK & VETCH C0. DUMMY
GILBERT CO. DUMMY

66 CO. DUMMY

COMMON C0. DUMMY

UNITED CO. DUMMY

BBEEEIBIENI * TANDA RROR

32.048
- .061
-8.720
.450

- .065
14.804
.705
2.049
.288
2.679
.135
.013

- .114
- .076
- .025

lllllllllllll
«b
O
N

34.596
.915
7.411
.467

 

* - Significant at 1 percent

+ - Significant at 5 percent

- - Significant at 10 percent
- .6974

Adjusted R2 - .4977

F(33, 51) - 3.492

 

81

Table 3-9. Two-Stage Least Squares Estimate Of the
Construction Cost Function

 

 

VARIABLE CQEEEICIENI SIAAOAAO_EAAQB
CONSTANT 20.391 40.221
PRESSURE DUMMY -3.431 4.877
LN(MW) -5.052 9.422
(LN(MN))2 .174 .632
-LN(1-EOUIV. AVAIL.) -13.332 9.194
(-LN(1-EOUIV. AVAIL))2 .730 .322+
LN(MW)(-LN(1-EOUIV. AVAIL)) 1.783 1.534
-LN(1-EQUIV. AVAIL.)*PRESSURE .412 1.122
PRESSURE*LN(MW) .459 .651
FIRST UNIT DUMMY .295 .069*
LN(YEAR) 2.582 1.950
COOLING TONER DUMMY .161 .100
FULL-INDOOR DESIGN DUMMY .015 .109
A-E SUBCRITICAL UNIT EXP. - .083 .184
A-E SUPERCRITICAL UNIT EXP. - .109 .121
MIDDLE ATLANTIC DUMMY .011 .208
NEST NORTH CENTRAL DUMMY .092 .253
EAST NORTH CENTRAL DUMMY .401 .342
ROCKY MOUNTAIN DUMMY - .021 .251
SOUTH ATLANTIC DUMMY - .127 .227
SOUTHERN SERVICES CO. DUMMY - .283 .215
TVA CO. DUMMY .227 .255
DUKE CO. DUMMY - .454 .280
AEP CO. DUMMY - .225 .584
STEARN G ROGER CO. DUMMY - .439 .388
BECHTEL CO. DUMMY - .143 .213
EBASCO CO. DUMMY - .275 .214
SARGENT G LUNDY CO. DUMMY - .098 .199
BRONN G ROOT CO. DUMMY - .295 .561
FLUOR CO. DUMMY - .528 .372
BLACK G VETCH CO. DUMMY - .338 .402
GILBERT C0. DUMMY - .164 .201
GC CO. DUMMY .002 .256
COMMON C0. DUMMY - .214 .407
UNITED C0. DUMMY - .439 .596

 

* - Significant at 1 Percent
+ - Significant at 5 Percent
- - Significant at 10 Percent
R2 - .6778

Adjusted R2 - .4543

F(34, 50) - 3.032

 

82

Table 3-10. Two-Stage Least Squares Estimate Of the
Reliability Function

 

 

VARIABLE BBEI’EIBIEEI _____E___$TANDARD RROR
CONSTANT 52.189 27.700-
LN(PER KN CONSTRUCTION COSTS) -12.142 6.236-
LN(MW) -9.472 4.442+
LN(MW)*LN (PER KN CONST. COSTS) 1.806 .974-
LN(YEAR) 2.868 1.730-
FIRST UNIT DUMMY .288 .113+
PRESSURE DUMMY - .162 1.947
PRESSURE*LN(MW) .008 .299
BOILER MANU. COMBUSTION ENG. .173 .131
TURBINE MANU. GENERAL ELEC. .317 .293
TURBINE MANU. WESTINGHOUSE - .004 .281
BOILER MANU. BABCOCK G NILCOX .190 .164
A-E SUBCRITICAL UNIT EXP. - .011 .152
A-E SUPERCRITICAL UNIT EXP. .007 .096
SOUTHERN SERVICES CO. DUMMY - .153 .184
AEP CO. DUMMY .400 .360
DUKE C0. DUMMY .383 .196-
TVA CO. DUMMY .237 .314
STEARN G ROGER C0. DUMMY - .262 .422
BECHTEL CO. DUMMY - .127 .158
EBASCO CO. DUMMY - .167 .144
SARGENT G LUNOY CO. DUMMY ~ .082 .159
BRONN G ROOT C0. DUMMY .009 .454
FLUOR c0. DUMMY - .831 .304*
BLACK G VETCH C0. DUMMY - .283 .259
GILBERT C0. DUMMY - .024 .179
COMMON CO. DUMMY - .285 .222
UNITED CO. DUMMY - .594 .310-
GC CO. DUMMY .523 .246+

 

* - Significant at 1 Percent
+ - Significant at 5 Percent
- - Significant at 10 Percent
R2 - .7819

Adjusted R2 - .6709

F(28, 56) - 7.043

 

83
considerably larger in absolute value and have a higher level Of
statistical significance than the OLS estimates of the coefficients.

The coefficient Of the size variable, ln(MW), has the expected
negative sign, but it is statistically significant in only one of the
four 2SLS estimates. The other size variable, (ln(MW))z, has the
expected positive Sign, but it is never statistically significant. In
general, the ZSLS estimates Of the coefficients Of ln(MW) and (ln(MW))2
are much larger in absolute value and are somewhat more Significant than
their OLS counterparts.

The time trend variable, ln(Year), always has a positive
coefficient and is significant in one of the four ZSLS estimates. The
corresponding OLS estimates Of the time trend coefficient are larger in
absolute value and have a higher level Of significance.

The first-unit variable has a positive coefficient that is
significant at the 1% level in all four 2SLS estimates. However, the
first-unit coefficients are quite Close to their values and significance
levels found in the OLS estimates. Also, several of the A-E dummy
variables reach some level Of significance and are quite similar to the
OLS parameter estimates.

Notice that the squared terms and interaction variables cause the
unit cost elasticities with respect to reliability and size to vary for
different locations in the sample. Unit cost elasticities with respect
tO size for the four 2SLS estimates are given in Tables 3-11 through 3-
16. Unit cost elasticities with respect to relaibility for the four
ZSLS estimates are given in Tables 3-17 through 3-22.73

Rom Roe Ran <m

com“ ofim "No 3:
amawmmm amawowm noun

“cocoon o~ pa acauLLL=oLm Y
AEGULGA m on Seaupewcmwm Y
pcmusoa a an acmuwwpcowm Y
oYm mpnah c, omugoamg mopzmos copmmmgoog no woman «a

e +-

 

84

mo¢.~ "Hm.~ amm.~ ~mm.~ ~5m.~ oom.~ oom.~ m~o.~ moo.~ omm.~ -e.~ h<hm h
+omo.m +mom.m +v~m.m +mmv.m +¢em.m +¢¢~.m +om_.m +ooo.~ +¢¢o.~ +¢mo.~ +o~v.~ Rmo
¢-.~ vo~.~ ¢-.~ mm~.~ mmN.~ os~.~ o-.~ ~m~.~ om~.~ moo.~ ago.“ h<hm h
+mom.~ +~m~.~ +mo~.~ +~mo.~ +~oo.~ +~om.~ +w-.~ +~¢o.H +~o¢.~ Ymom.m mmo._ goo

o~m.~ w~m.~ oom.~ oom.~ omm.H moo.“ m~o.~ -N.H vfim.~ H-.H mam. h<hm h
Yogm.~ Ym¢¢.~ Ymnm.~ Y~o~.~ Y~o~.~ YfioH.H ono. Yumm. Nos. Nam. mom. Rom
mom." mom.— mom.“ ¢¢M._ mom.~ om~.~ woo.” woo. mfiv. on“. Y moo. Y h<hm h
mmo. mom. Ham. omn. ovo. ovm. “No. now. He“. ovo. Y mow. Y Rom
mom. ave. mus. mum. mom. on. omo. Y mum. Y evo.~Y svm.~Y ¢¢¢.HY h<hm h
mam. omv. osm. mow. mom. mo“. moo. Y on“. Y cow. Y owo. Y om“. Y Ros
own. vou. ovo. meg. Y Kso. Y Nmo. Y ~vm.~- ~o~.~Y oo~.~Y o-.~Y “mo._Y h<hm h
Nag. voo. owo. ~oo. Y om“. Y omm. Y oom. Y +mo¢. Y +omo. Y +oem. Y vao.~Y Ron
~om.Y soc. Y on“. Y oHH._Y oHo.HY ¢m~.~Y moo.~Y o~H.mY coo.~- ~mo.~Y mmm.~Y h<hm h
mm~.Y mow. Y oww. Y goo. Y Hmv. Y +~mm. Y «woo. Y «mos. Y eomo. Y +ov_.~Y +mwm.~Y xmo

oNo.Y oo~.~Y ~mm.~- ego.~- ~om.~- _-.mY omo.mY ouo.mY mam.mY «Ho.~Y mom.~- p<~m h
oom.- ~o¢.Y “on. Y +-o. Y +Hfis. Y sggo. Y smwo. Y «omo.HY 8H-.~- «oo¢.~Y +m¢o.HY goo
.Fwa>< .>L=am

 

 

can" ooum oo—— coo" coo coo can com com cov con
whh<x<amx

 

3.2.5 6:28 a 28 2:» 3
pounds: ES.: us.».u.uaa_u Smog as... .HA-n 6.5.L

 

85

xmo Rue
mos on
amaflmma amaflofla

acmugma o“ um acmuLoL=oLm u
acmogmq o as acmupmwcoLm Y
acmugoa a an «cauwopcowm n

xon <o
moo 3:
aqua

'84-

qu «Pomp cw omugoamg uppzmms cowmmosomg co ommmm as

 

 

 

one.~ mm¢.~ -¢.~ o~m.~ oom.~ o¢H.~ p<pm h
+~¢o.m +o¢o.m +~m¢.m +oom.m +~m~.m +o~o.~ goo

~o~.~ ¢o~.~ ¢o~.~ ooo.~ coo." mom.“ p<hm H
+o~o.~ +~oo. +ooo.~ +~ms.~ Yoso.~ mom." goo

mos.“ ¢-.H oos. ~mo.~ oo~.~ ofik. h<ho »
Yvon.“ Ynoo. -ooo. -o. ooo. ooe. aoo

coo." goo. ooo. ouc. nwo. mom. Y p<hm p
ofio. ogo. oom. on". "Ho. om“. Y goo

umo. cow. Y Koo. Y «mo.HY vo~.~- oo~.~Y ~<pm h
woo. moo. Y ~o_. Y omm. Y Nov. Y Nos. Y Ros
ovo.~- ~o¢.~- Noo.~Y ooH.~Y ~o~.~Y ooo.~Y ~<hm h
mom. Y oov. Y Yooo. +~¢~. Y +moo. Y YmH~.~Y aon
m-.~Y H-.~Y Noo.~Y omo.~Y ~oo.~- om~.~Y h<hm h
-ovs. Y +mvo. +ooo. +ooo.~- +~o~.~Y +oo¢.~Y goo
¢o~.NY ooo.~- ¢¢~.~Y oHo.~Y Noo.~Y omv.~Y h<hm h
+s¢o.~Y +o¢~.~Y «~o~._Y «oom.~- s~oo.~- +~o~.~Y goo

.P_~>< .>.=ao
ooo oos ooo ooo oov oon
mhh<3<oox

 

«neon: —nowu—soa=o a so» onpm ou
vooomoz sup: mopuoowumn—u «moo nan—m .«uYn o—aah

 

86

Nam wpnmh cw UmHY-Oam.» 9:5me ccrwmmhmw.» :O tmmmm

Rwo

oomH
amawmma

goo goo <o
mom mug ::
amawqfla omwa

ucwugma oH «a unaupwwco_m u
acousma m we acauwmwcopm Y
acmugma a “a Hemopomcopm Y

*4-

 

 

 

 

oo~.~ mo~.~ oo~.~ ~o~.~ ~¢~.~ m-.~ go_.~ oo~.~ ~o~.~ ~Ho.~ mgo.~ h<hm »
+on. o +~m~. o +ooH.¢ +o-. o +m¢o.¢ +gom.m +ooo.m +~¢g.m +m~o.m +~ov.m Ym¢~.m goo
omo. ~ omo. ~ oNo.~ moo. N Hoo.~ “so.“ mum.“ moo." oog.~ woo.“ mom.“ h<hm h
+geg. ~ +ooo. u +o~o.~ +ooo. ~ Yo~¢.~ Yeom.~ Ygo~.~ Yeo~.~ Ymoo.~ moo." coo.“ goo
sou.“ oog.~ mos.“ gag.“ mos.“ ooo.~ moo.“ -¢.~ “on." ogo.~ vgg. h<hm »
Ymmo._ Yogg.~ Y-5.H Yuvo.H Yooo.~ ago.“ ~om.~ og~.~ gmH.~ ogo. ooh. goo
omv. m o~v.~ oom.~ son." ofim.H om~.~ o-.~ goo. mwg. moo. gm“. p<hm p
vo_. a o-.~ moo._ moo. ooo. “no. nos. ~No. moo. on. gag. goo
No9 A ogo. o~o. ~oo. ogg. oco. oov. NMN. “No. Y ogu. Y moo. Y h<hm h
floo. m~o. ooo. ooo. mac. gum. omw. ogo. o~o. Y “so. son. Y gog
oov. oom. oom. mug. moo. mom. Y goo. Y oog. Y moo. Y ¢o9 "Y ooo._Y h<hm h
ogm. -~. ova. ago. woo. ¢o9 goH. Y mow. Y owe. moo. Y was. Y gog
to“. Y oom. Y msv. Y mos. Y goo. Y Non. HY ogo._Y g~g.~Y o_g. "Y omo._Y m~o.~Y p<hm h
og9 Y on“. Y om“. Y mow. oem. “no. Y oNo. Y Yfivo. Y Ymgg. Y omo. Y oe_.~Y goo
woo. Y omo._Y oom.~Y oeo. HY o¢9 NY oo¢.~Y moo.~- ooo.~Y oom.~- o-.~Y Hoo.~Y h<hm »
og9 Y «no. Y goo. Y og9 Y +g¢9 Y +~mg. Y +omo. Y +~¢o. Y +ogo.HY +gm~._Y Yo¢¢.~Y goo
.—Pa>< .>L:oo
oon— oouu cog“ ooo— ooo ooo oos ooo ooo cot con
mph<=<oox
Ge u.:= —uu,u—Loeon:o a Low
onpo on vacuums :u.: movupuwumnpo umoo «cup; .mfiYn opnnh

 

87

oYo «Pomp co umpgccmg mapcmms coommmgmmg cc commo

goo

oooH

amawmmn

ucoogmc co an ucmuooocmpo Y

goo

oHo
amaflofln

gag
Moo
noun

<o
3:

acmugmc o pa acauoopcmwo Y +
acougmg H on acmuvopcmoo Y 4

 

 

 

 

ooo." o-.~ ooo.“ ooo." ooo.“ om~.~ oo~._ ooo.“ ooo.“ ooo." ooo." o<po h
ogo.m ooo.m o-.o ooo.o ooo.m goo.o goo.o o-.o oHc.o o~o.~ ooo.~ gom
oc~.~ mc~.~ mcH._ ooo.“ moo.“ goo." ooo.“ omc.~ floc.~ moo.“ ogm. ~<ho p
ooo.~ Hoo.~ oco.~ g_~.~ -~.~ ooc.~ omo._ oog._ omo.~ omo.~ oo~.~ gcm
ggm.c ooo. mom. ooo. mom. com. com. ooo. ooo. ocg. moo. h<ho h
ooo.“ coo._ «so.~ oom.~ ~m~.~ ooH.~ ooo." oom. Non. ooo. ooo. goo
c~o.o “mg. cog. oog. oog. ooo. coo. c_o. "Ho. moc. Y goo. Y p<ho h
ooo.“ com. ooo. oog. ocg. omo. ogo. goo. Nu". ooo. Y goo. Y goo
ooo.c moo. oco. goo. ooo. moo. soc. “on. Y coo. Y ooo.“- ooo.~Y p<ho »
ogo.c moo. ooo. ooo. ooo. ooo. ooo. Hog. Y ooo. Y ooo. Y oog. Y gog
oo~.c ooo. ooo. ooo. Y ooo. Y ooo. Y ooo.”- ooo.“- ooH.~Y ooo.“- oco.HY p<ho p
~c~.c moo. Hoc. ooo. Y on“. Y ooo. Y ooo. Y Yomo. Y +moo. Y Ycoo. Y ooo.“- gcg
~o~.cY ooo. Y Ago. Y ooo. Y ooo._Y mcH.~Y mcg.~Y ooo.~Y oo~.~Y Hom.HY ooo._Y h<ho h
o_~.cY goo. Y ooo. Y "on. Y ooo. Y +~oo. Y Yugo. Y +-o. Y +ogm. Y Yog~.~Y ooo.“- goo
cco.cY ogc.~Y moo.~Y oom.~Y ooo.~Y coo.~Y oog.~- ~go.~- co~.~Y mmo.~Y ogo.~Y p<po h
ooo.cY ooo. Y moo. Y Yooo. Y +mog. Y «ooo. Y Yoom. Y +ooc.~Y +oo~.~Y Ymoo.~Y Yocg.~Y goo

.P_~>< .>_cao
coo“ coon ccao coco coo cco cog coo coo coo cco

th<3<oox
e8 pogo oococoo a so» ¢~.o cu
noocmoo coo: mo—u—ooumupo umoo ago—c .oHYo o—cnh

 

gmm ggo gog <o
oog com moo z:
amawmmn amaoqon moon

“coupon co co acmuoooomoo Y Y

nomugmo o um acmogowcmoo Y +

pomucmo g on ucaupoooooo Y «
mYm mpcah op cougcomg magnum; ocpmmmcmme cc ommao ««

 

88

 

 

moo. ooo. mgo. gom. oom. ooo.“ o<po o
o~o.o moo.o ooo.o ooo.o ogm.o ogo.o gom
ooo. moo. ooo. omg. oog. mmo. p<po h
oom.~ mmo.~ coo._ o~o.~ mm_.~ omo.“ gom
goo. mmo. moo. ooo. ogo. ooo. o<ho h
goo. coo. goo. omo. ooo. oom. goo
go". mmc. moo. omo. Y goo. Y mom. Y o<go h
ooo. gmc. ooo. coo. Y goo. Y gmH. Y goo
ocm. Y omo. Y coo. Y ooo.“- ooo.“- oom. Y g<oo h
ooo. Y com. Y ooo. Y ooo. Y omo. Y omo. Y gog
mom. Y ooo.“- gog.~Y o_~.oY ooo._Y ooo.“- o<oo o
mgo. Y ooo. Y Yogo. Y +mog. Y Ycoo. Y com. Y gcg
omo."- ~o~.o- ooo.o- o_o.oY ogo.~Y omo.“- o<ho o
ooo. Y +ccm. Y +oom. Y +g~c.~Y omo."- oo~.HY goo
ggo.oY ooo.oY moo.oY ooc.oY ooo.“- ooo.“- p<oo h
+omc.~Y +om~.~Y +omH.~Y +ooo.HY mmm._Y moo.~- goo
.gpm>< .>L=oo
coo ccg coo coo coo ccm
ohh<3<oox

 

t. :5 32:93 a as on; 3
888.. 5:. 85.63:; :8 as: .23.. as:

 

89

goo
oomH

gmo
mom

ago-nag

Homogmo oH Hm HcauHoHcmHo Y

goo
mgg
an

<o
3:

acousoo o Ha HoauHoHcmHo Y +
chugmc H on HcauHoHomHo Y «
mYm chah oH omogcomc mHHcmmL :chmmmeg co ommao «5

 

 

 

 

Hmm. oom. ooo. com. Hoc.H ooc.H moo.H goc.H ooH.H ogH.H goo.H h<oo h
ooo.m oHo.m ooH. m oo9 m oH9 m ogc. m ooc.m ogm.o HHm.o mmo.o mmg.o gom
ooo. moo. moo. ooo. mHm. oom. Hom. moc.H ogc.H mHH.H coH.H o<oo o
ooo.o com.H co9 H on. H ooo. H mmo. H omg.H mmg.H ogo.H omo.H gmo.H gom
mcg. oog. oog. mog. om9 omo. Hgo. on. Hom. com. Hog. h<oo h
ooo.H goo.H goo.H mm9 H go9 H oH9 H moc.H oHo.H oom. ogo. ogg. goo
coo. ooo. moo. mco. oo9 ooo. ogo. ooo. mmo. ooo. mgo. o<ho h
ogg. oog. ng. Hoo. ooo. moo. ooo. moo. mmo. oom. ooo. goo
oom. gom. mo9 mom. oom. Hmm. Hmo. moo. ogc. moc. omH. Y h<ho h
ogm. oom. on. moo. ooo. ooo. moH. ocH. Hoo. omo. omH. Y gog
ooo. ooo. oH9 Y og9 Y mo9 Y ooo. Y oHo. Y ogo. Y ooo. Y ooo. Hmm. Y h<oo p
moo. Hoo. mc9 Y ooo. Y mg9 Y coH. Y goH. Y ooo. Y ooo. Y Hom. Hoo. Y gcg
oo9 Y moo. Y coo. Y go9 Y oom. Y goH. HY Hgo.HY HHm. Y mog. Y coo. moo. Y h<oo o
oo9 Y ooo. Y ooo. Y gHm. Y oom. Y omm. Hoo. Y omo. Y ooo. Y omo. omg. Y goo
ogo.Y ogc.HY on. HY ooo. HY Hgo.HY moo. HY on.HY ooc.HY ocm. Y oog. ooo. Y o<oo o
ooo. omo. Y ooo. ooo. Y omo. Y omo. Y mgo. Y mmg. Y gmg. Y ogo. ogm. Y goo

.HH~>< .>H=oo
ccmH ocoH ocHH cocH com coo cog coo coo coo com

ohh<3<oox
«e uHco HuoHuHLosooco a com oNHo ou
accomoo :uH: moHuHoHumoHo umoo non—o .oHYm oHooh

 

90

gmm
oomH

 

scemez

goo

on
EceHcHz

 

gHg
Hoo

moms

(o
3:

Hcmugwo oH Hm HomoHoHcmHo Y Y
Homugmo o Hm HomuHoocmHo Y +
Hcmugmo H on HomuHoocmHo Y a

 

 

 

oYm chap co omogcomg oHHcmmL ccHoomcooL cc ommao ««
mmg.o Hgg.o oog.o mHg.o mgo.o Hoo.o Hoo.o ooo.o mom.o ooo.o ooo.H p<po »
«oom.m «omo.m «ooo.m «ogo.m +Hgo.m +Hoo.m +Hog.o +ooo.o +ooH.o +moo.H ooH.H gom
HHo.o gog.o oog.o gHg.o goo.o goo.o omo.o mmm.o ooo.o gmo.H Hgm. p<oo o
chc.m Rooo.o «ogo.o «moo.o +ooo.o +ooc.o +mmg.H +mmo.H +gmH.H oog. HoH. gcm
mog.o oog.o mog.o ogo.o moo.o mHo.o oom.o ocH.o ooo.H coo. omo.HY H<ho H
«moo.o «goo.o «goo.o +on.H +ocg. +ogo.H +oHo.H +on. coo. ooH. omo. Y goo
gog.o ooo.o mmo.o ogo.o ogo.o Hmm.o moc.o ooo.H moo. Hom.HY goo.oY h<ho h
«oHc.o «goo.H +ooo.H +ooo.H +gmo.H +goc.H +oco. ooo. coH. ooo. Y «ooo. Y goo
ooo.o ooo.o Hmo.o mom.o omo.o mmm. mgo.H mog. omg. Y ogo.oY moH.mY h<ho o
+omo.H +ooo.H +ogm.H +ooH.H +mgm. Ymog. moo. ooH. goH. Y +mco. Y «ooH.H- gog
ooo.o mom.o ogo.o oHH.o ggo. ooo.H mmo. oom. Y omg.HY ooo.o- oHo.mY h<ho g
+omo.H +coo.H +cHH.H +oom. YmHg. moo. moo. ogo. Y Ygoo. Y «ooo. Y «moo.H- gcg
ooo.o goH.o oom.H oog.H moo.H ogo. omo. oHH.HY omo.oY mHm.oY mmH.mY h<ho o
+gHo.H +Hoc.H YHmo. Yocg. ooo. cgo. oHc. Hmo. Y +goo. Y «ooc.HY «moo.H- goo
ooo.o ooo.H ooo.H omm.H omo. ooo. mHo. Y omo.HY ooo.o- oom.oY HoH.mY h<oo p
+goc.H Yogo. Hog. oHo. mom. ooo. HoH. Y Hoo. Y +gmo. Y «ogo.HY «omo.HY goo
.HHa>< .>H=ao
ccmH cooH ccHH cccH ccm cco cog cco coo coo ccm
ophoxoou:

 

6» auaamog gqu maHHHuHum._o «Goo pea—a

t :5 6:28 a 6.. 3:223.

.gHYm o—coh

 

91

Rom
oomH

3%

6%
Sn
3...: ..

acmugwo oH an “coupopcmwm u

RH“

Hoo
anon...

<m
32

 

 

 

 

acmugao o on ocmuwopcmpo a +
acmugmo H on acmupowcowm u «
oYo mHoah =_ omusoomg maHommg ocwmmmgmms co ommao *«
ooo.o Hoo.o Hoo.o oHo.o mom.o ooo.o moo.o ooo.o ooH.o ooo.H ooo.H h<pm »
+ooo. o +ooH.o +oom.o +mHo.o +ooo.m +oHo.o +mom.o +Hoo.o +ooH.o Yooo.H omm. gom
ooo. o ooo.o oHo.o omo.o ooo.o oHo.o ooo.o Hmo.o oHo.o ooo.H ooo. h<ho »
+ooo. o +ooH.m +mom.o +ooo.o +oHo.o +Hoo.o +mom.H +ooo.H +HmH.H ooo. moo. gom
moo. o ooo.o ooo.o omo.o ooo.o omo.o omo.o omo.o ooo.H ooo. ooo.H- h<po h
+ooo. o +ooo. o +ooo.o +omH.o +mom.H +moo.H +ooo.H +Hoo.H ooo. ooH. ooo. Y goo
ooo.o ooo. o ooo.o ooo.o ooo.o moH.o ooo.o ooo.H moo. mmH.HY ooo.o- p<ho h
+ooo.o +ooo. o +moo. o +omo.H +ooo.H +Hmo.H Yomm. ooo. Hmo. ooo. Y +Hom. Y goo
ooo.o Hoo.o ooH. o ooH.o Hoo.o moo.H ooo.H Hmm. ooo. Y ooo.o- mom.o- p<po h
+ooo.o +omo.H +Hoo. H +ooo. H Yooo. H Yoom. Hoo. ooo. ooo. +Hoo. Y «moo.H- goo
omo.o ooo.o oom. H ooo. H ooo. H ooo. H ooo.H ooo. ooH. HY ooo.o- ooo.o- h<ho h
+ooo. H +ooo.H Yooo. H Yooo. H Yomm. ooo. moo. Hoo. ooo. Y +moo. Y «ooo.H- goo
oom. H ooo.H oH5 H moo.H ooo. H Hoo. H ooo. ooo. Y omo.HY ooo.o- oHo.mY p<ho »
YoHo. H YHmo.H Yomo. H ooo.H oo5 oHo. oHo. ooH. Y ooo. Y +ooo.HY «omo.HY goo
Hoo.H oHo.H ooo. H omo.H oo9 H oo9 moo. ooo. Y ooo.H- ooo.o- oom.oY p<ho h
Yooo.H ooo.H ooo.H ooo. omo. omo. Hoo. oHo. Y Yooo. Y +Hoo.HY «ooo.H- goo
.Ho~>< .>F:oo
oooH oooH ooHH oooH oom ooo ooo ooo ooo ooo ooo
ohh<3<ooz
a. non: upoccco o ooo ooo—HooHHoo on
vacuums zaHz moHuHoHumoHu umoo uono .oHYo oHooh

 

goo goo goo <m

ooh on moo 3x
amaflmmm amaflofla mama

acousmo oH an acauooHcmHm a Y

ucmugwo o no acmuwopcowm a +

ocwugmo H on ocmuwowcmpm u a
oYm mHnmh :_ umpgoomg muHomms copmmmgmmg co ommno ««

92

 

 

 

ooo.o ooo.o ooo.o omo.o ooo.H oHo.H h<hm h
+moH.m +ooo.o +ooo.o +ooH.o Yoo~.H oHH.H goo

Hmo.H oom.H oom.H moo.H moo.H ooo. h<hm h
YooH.o Yooo.H Yooo.H YooH.H ooo. ooH. goo

ooo.H ooo.H ooo.H ooo.H moo. ooo. Y h<hm h
oom.H Hom.H oom. HHo. ooH. ooo. Y goo

oHH. ooo.H Hoo. ooo. omo.HY ooo.o- h<hm h
ooH.H moo. moo. moo. ooo. Y +ooo. Y goo

oHo. ooo. omm. ooo. Y ooo.o- oom.oY h<hm h
ooo. ooo. ooo. moH. Y +ooo. Y +noH.HY goo

moo. ooo. oHo. ooo. Y ooo.o- ooo.o- h<hm h
omo. moo. ooo. ooo. Y +ooo. Y «oHo.H- goo

oom. oHH. ooo. Y ooc.HY omo.o- ooo.o- h<hm h
oom. ooH. oHo. Y ooo. Y +Hoo.HY «Hmo.HY goo

HoH. ooo. Y oHo. Y omo.HY mom.o- oom.oY h<hm h

HmH. ooo. Y omm. Y ooo. Y +ooo.HY «oHo.H- goo

.oH<>< .>H=cm
can can ooo ooo ooo oom
mhh<3<amx

 

».== .au.u.ooa=m . ooo ooH....._¢¢ on

goofing: gu.z «o.»_o.um._o "moo ago—g

.mHYn o—ooh

 

93

goo
oooH
amaflmmn

goo
mom
amaflqou

acousmo oH an ucnuHopcoHo u
acmugmo o pa acauomwcoom u
acmugmo H on pcauHoHcoHo u

goo

ooo
aqua

<m
3:

« -+

oYo «Hoop or ompgoowg moHommg conmmgmwg cc ommao ««

 

 

 

Hoo.o omo.o ooo.o omo.o ooo.o ooo.o ooo.o HoH.o ooo.H ooo.H ooo. h<~o p
aooo.o «omo.o *HH5 o «oHo.o +ooo. o +ooo.o +moo.o +moo.o Yomo.o ooo.H ooo.H gom
oHo.o ooo.o oH9 o ooo.o ooo. o mom.o ooo.o oom.H ooo.H ooo. ooo. h<ho h
+ooo.o +on.o +oo5 o +moo. o +ooo. o +Hoo.o +ooo.H YHmo.H oHH.H ooo. ooo. gom
omo.o Hoo.o oHo. o ooo. o ooo.o ooo.o omo.o ooo.H ooo. HoH. oHo. Y p<po p
+moo.o +Hoo.o +ooH. o +oom. H +Hoo.H +moo.H +ooo.H Yoom. ooo. ooo. Hoo. Y goo
ooo.o omo.o moH.o omo.o mom.o ooo.o ooo.o ooo.H ooo. omo. Y ooo. Y h<hm p
+mmo. o +mom.H +ooo.H +Hoo. H +ooo. H +ooH.H +ooo. oHo. ooH. oHo. Y oom. Y goo
mm5 H moo. H ooo.H on. H oom. H oom.H ooo.H omo. ooo. Y oHo. Y ooo.H- h<hm h
Yooo.H Yoo9 H Yooo.H Yooo.H YHo9 H Ymoo. oHo. mmH. ooH. Y ooo. Y ooo.H- goo
ooo.H oo9 H ooo.H oo9 H Ho9 H ooo.H oom. moH. Y omo. Y ooH.HY ooo.H- p<ho h
ooo.H oo9 H ooH.H mom. oo5 oo9 ooo. ooo. Y Hoo. Y moo. Y moo.H- goo
ooo.H Ho9 H ooo. H ooo. H oHH. H ooo. ooH. Hmo. Y ooo.H- ooo.H- ooo.H- h<bo »
oHo.H ooH.H oom. oo5 ooo. oHo. Hoo. ooo. Y moo. Y ooH.H- omo.HY goo
ooH.H gmo.H ooo. H on. oHo. ooo. ooo. omo.HY ooo.H- moo.H- ooo.H- p<ho »
ooH.H oom. moo. ooo. ooo. ooH. ooH. ooo. Y ooo. Y omo.HY ooo.H- goo
.HH~>< .>H=oo
oooH oooH ooHH oooH oom ooo ooo ooo ooo ooo ooo
ohh<3<omx

 

.. non: .au.»..o.¢a=m . Loo a»._.a~..o¢ on
goofing: ;¢.z m¢.u.o_ua._u page ecu—o

You-n oponh

 

goo goo goo <m
oom on moo 3:
amaflmma amaflqfla moms

acmugmo oH um ucaumecoHo n Y

acmugmo o um acmuHchmPo u +

ocmugmo H on acauHoPcmHo - «
oYo oHoah :H omugoomg moHomwg conmmgows co ommao ««

 

94

 

 

Hoo.o mmo.o ooH.o ooH.o mmo.H ooo.H h<ho h
+oom.o +ooo.o +moo.o +ooH.o Yooo.H oHo.H gom

ooo.H ooo. omo.H ooo.H ooo.H ooo. h<ho h
oom.H oHo.H Ygoo.H YoHH.H oHo. ooo. gom

ooH. ooH.H HoH.H ooo.H oHo. ooo. Y h<ho P
ooo.H ooH.H ooo. ooo. ooH. omo. Y goo

oHo. omo. ooo. oHo. omH.HY ooo.H- ~<po »
oom. ooo. ooo. ooH. omo. Y YoHo. Y goo

ooo. ooo. ooH. ooo. Y ooH.oY omo.o- h<po »
oHo. ooo. ooH. ooo. Y +ooo. Y +ooH.HY goo

omo. moH. oHo. omo. Y oom.oY ooo.o- h<ho h
ooo. ooH. ooH. omo. Y +moo. Y +ooo.HY goo

ooH. oHH. Y ooo. ooH.H- ooo.o- oog.o- h<ho »
HoH. oHH. Y omo. oHo. Y +oHH.HY «ooo.H- goo

moo. omo. Y Hoo. ooo.H- ooo.o- ooo.o- o<ho h
ooo. oHo. Y ooo. on. Y +moo.HY «ooo.H- goo

.H_m>< .>w=co
ooo ooo ooo ooo ooo ooo
ohh<3<ooz

 

c¢ u—c: Havoc—Luooo a Low huopooooch on

888.. 5:. 833325 :8 as: 4an oz:

 

95

goo
oooH

amaflmmm

goo
moo
azswmmn

ucmosmo oH am acouochmHo a

goo

moo
moon

<m
3:

 

 

 

 

acouswo o an ucmuonoowo u +
acmugmo H on ucauHoHcmwo u «
mYo mHoah co omuocoms maHommg ooommmgowg co ommmo ««
HHo.o ooo.o ooo.o mmo.o ooo.o mHo.o ooo.o HoH.o oo5 H oo9 H ooo. ~<po h
+ooo. o +gmo. o +ooo. o +og5 o +ooo.o +ooo.o +ooH.o +Hoo. o Yomo. o moH. o ooo.H om
mo9 o oHo. o Hoo. o omo. o ooo.o HHo.o ooo.o oom. H ooo. H ooo. ooo. h<ho p
+oo9 o +oo9 o +oom. o +oo5 o +ooo.o +ooo.o +ooH.o Ymoo. H ooo.H ooH.H oHo. om
ooo.o oHo.o ooo.o moo.o moo.o oHo.o ooH.o Hoo. H oo9 H ooo. oHo. h<ho h
+ooo. o +om9 o +oo9 o +ooH.o +oom.H +ogo.H +ooo.H ooo. H oo9 ooo. ooo. oo
gH9 o ooo. o HoH. o Hoo.o ooo.o ooo.o on.H mH9 H moo. moH. goo. Y h<po h
+oHo. o +oo9 o +on. H +ooo.H +ooo. H +oo9 H YoHH.H ooo. oHo. oHH. omm. Y oo
Hoo.H moo. H ooo.H oom.H oom. H moo. H moo.H mmo. ooo. ooo. Y oHo. Y p<ho h
Yomo. H Yoo5 H Yomo.H Yooo.H Yomo.H Yoo9 H ooo. oHo. goH. HHo. Y ooo. Y oo
oo9 H Hoo.H ooo.H moo.H oHo.H ooo. H omo.H ooo. HHH. Y moo. Y ooo. Y h<ho »
oo9 H Hoo.H ooo.H ooH. H oom. oo5 ooo. ooo. oo9 Y oo9 Y omm. Y oo
oo9 H oo9 H oo9 H oo9 H ooo. H ooo.H ooo. omo. moo. Y oo5 Y moo. Y p<ho h
mm9 H oo9 H HoH. H Hom. oo5 ooo. omo. ooo. ooo. Y ooo. Y oHo.H- oo
ooH.H ooH.H mm9 H ooo.H on. ooo. oHo. ooo. omo. Y omo. Y ooo.H- p<po h
ooo.H Hoo.H oom. omo. ooo. ooo. ooH. ooH. ooo. Y omo. Y oHo.H- oo
.HP~>< .>H=oo
oooH oooH ooHH oooH oOm ooo ooo ooo ooo ooo ooo
ohh<x<oos
at uHco Hoo.o—Luoooao a so» ou—HHooHHco 0»
«cacao: guH: moHuHoHumoHo umoo ecu—o .ooYo oHooh

 

96

An examination of the unit cost elasticities with respect to size
reveals several things of interest. First, the largest degree of
statistical significance occurs at low levels of reliability. Also, the
elasticities are always negative at low levels of reliability and they
diminish in absolute value as unit size increases. The elasticities are
always positive at higher levels of reliability and increase as unit
size increases. This suggests that per Kw construction costs can be
reduced if unit size is increased at a low level of reliability, but
that maintaining a high level of reliability as unit size is increased
will cause higher per KH construction cost.

The unit cost elasticities with respect to reliability are
unexpectedly negative and take on some level of significance for the
smaller sized units. As unit size increases, the elasticities are
negative only at low levels of reliability. Generally, the elasticities
for units beyond 600 NH or 700 MW are all positive and take on a fairly
high level of significance. Notice for any particuluar level of
reliability that the elasticities tend to increase as the units get
larger. This suggests that the impact of reliability on unit
construction cost is stronger for large generating units regardless of
whether they are of subcritical or supercritical technology.

An examination of the ZSLS estimate of the values for the
reliability equation (see Table 3-10) reveals several things of
interest." First, the time trend variable, ln(year), has a positive
and statistically significant coefficient. This seems to indicate that
units that entered commercial operation later in the observed period had

higher levels of reliability than similar units that entered commercial

97
operation earlier. The ZSLS estimate of the time trend coefficient is
twice as large and has a higher level of significance than the OLS
estimate.

The unit-size variable, ln(MH), has a negative and statistically
significant coefficient. Interestingly, there is a negative and
significant coefficient for the average construction cost variable.
Also, the interaction between unit size and average construction cost
has a positive and significant coefficient. It is important to note
that the OLS estimate of these coefficients is considerably smaller in
absolute value than the ZSLS estimates. However, the OLS estimate of
each of the coefficients has a higher level of statistical significance
than the ZSLS estimate.

Several of the A-E dummy variables reach some level of statistical
significance and are similar to the OLS parameter estimates. Again, it
is curious that the first-unit dummy variable has a positive and

significant coefficient.

m:

The purpose of this chapter has been to develop a model that
recognizes reliability as an endogenous attribute of baseload coal-fired
generation. The capital-intensive nature of baseload coal-fired
generation means that intensive utilization can significantly reduce
average capital cost per KWH generated and thus average total generation
costs per KHH generated.

He began this chapter by noting that the cyclical demand for
electricity means that generating units will be designed for different

98
modes of operation. For example, baseload units are designed for
continuous operation at maximum capacity whenever they are available
while, at the other extreme, peakers are designed for rapid starts and
shutdowns and are operated only during periods of peak demand. In order
to avoid the problem of mixing units designed for different modes of
operation, we restricted ourselves to an analysis of the generation
costs of coal-fired baseload units.

An examination of the unit design process revealed that the
fuel-output relationship (we ignore labor) is fixed once the unit is
built and is fairly insensitive to the level of generation. Thus, the
utility is faced at the design stage with choosing unit engineering
attributes, such as size, thermal efficiency (as measured by steam
pressure conditions), and reliability so as to minimize the expected
total cost of meeting the expected load.

Given the capital-intensive nature of baseload generation and the
putty-clay nature of the technology, an important determinant of a
unit’s total generation cost per KWH is the intensity with which it is
used. Supercritical units are slightly more fuel efficient than
subcritical units with steam pressures of 2400 PSI, but they are also
generally more costly to build and less reliable. So we concentrated on
developing a unit average construction cost model that would allow us to
examine how capital cost per KN varies for the two technologies given
various combinations of unit size and reliability.

The result was a simultaneous-equations model where construction
cost per KH and reliability are endogenous variables. A translog form

of the construction cost function was estimated with unit size,

99
reliability (as measured by equivalent availability), steam pressure
conditions (subcritical or supercritical), first unit at a multiunit
site, and A-E experience with the technology as the primary explanatory
variables. The reliability equation was estimated with unit size, steam
pressure conditions, and the quality of the facility as measured by
capital cost per KH as the primary explanatory variables.

The chapter concluded with a review of OLS and ZSLS estimates of
the average construction cost and unit reliability functions. The above
results suggest that the potential bias from ignoring the simultaneous
nature of the model and/or ignoring the measurement error associated
with the use of realized reliability instead of the gx_agtg level of
reliability may be important factors in the estimation of generating
unit construction cost-unit reliability relationships.

In Chapter 4, we will show, assuming that whenever a unit is
available it is used, how annual capital costs per KWH change for
various combinations of unit size and reliability for both subcritical

and supercritical technologies.

CHAPTER 4

AVERAGE CAPITAL COSTS AND THE UNIT SIZE-RELIABILITY TRADEOFF

W

In Chapter 3, we developed and estimated a simultaneous-equation
model with construction cost per KH and reliability being treated as
endogenous variables. We will use that model in this chapter to show
how capital costs per xv and capital costs per KHH respond to various
combinations of unit size and reliability for both subcritical and
supercritical generation technologies. These results are interesting in
light of a number of existing and proposed programs aimed at improving
the performance of the electric utility industry.

The cost-plus nature of electric utility regulation has long been
suspected of reducing incentives to make efficient investment and
operating decisions by utility management{” So, in an effort to
improve the efficiency of utility operations, a number of state
regulatory commissions have implemented incentive programs which
condition financial rewards or penalties on some measure of a utility’s
performance. A popular incentive program involves setting generating
unit performance targets. The most common criteria used to measure unit
performance is equivalent availability. But setting minimum standards
for a narrow range of performance criteria can create perverse
incentives for utility management. For example, utility management
might be willing to incur excessive costs in areas outside the incentive

program so as to improve performance in the targeted areas. Thus, it is

100

101
important to understand the extent to which there is a trade-off between
the reliability and the construction costs of a generating unit.

The Federal Energy Regulatory Commission (FERC) has recently
proposed all-source bidding programs for new generating capacity in an
effort to improve the efficiency of utility investment decisions. A
major concern under any all-source bidding program is how to evaluate
and weigh the price and nonprice characteristics of each proposal when
ranking the bids. An important attribute to include when evaluating a
bid for baseload generation capacity is the availability or reliability
of the proposed unit. Each unit design is likely to have a different
level of reliability. Thus, it is important to determine the
consistency between the units’ design, projected construction costs, and

expected availability.

MW

Our primary purpose at this stage is to evaluate how capital costs
per KH and capital costs per KHH respond to variations in unit size and
reliability. An examination of both is necessary because the cost-
minimizing level of reliability for a given sized unit often differs
between the two measures of capital costs. This analysis will be done
by using the average construction cost function estimates given in
Tables 3-6 through 3-9 in Chapter 3. In order to derive estimates of
capital costs per KHH, we must make the assumption that a generating
unit is used whenever it is available. This is a reasonable assumption

given that all units in our data set are baseload in nature.

102

The results of this exercise are presented in Tables 4-1 through 4-
14. The odd-numbered tables (4-1, 4-3, etc.) show how capital costs per
KWH respond to variations in unit size and reliability, while the even-
numbered tables (4-2, 4-4, etc.) show how capital costs per KH respond.
Each table was derived by assuming a unit was either subcritical or
supercritical, the architect-engineer had previously designed and built
five generation units with the given technology, the presence of a
cooling tower, a full-indoor design, the unit is not the first unit at
the plant site, an annual fixed charge rate of 15 percent, and that the
unit entered commercial operation in 1970.

A general result for each measure of capital costs is that the
cost-minimizing level of reliability falls as unit size increases. This
is consistent with the observation in Chapter 3 that high levels of
reliability are more expensive to attain for large units regardless of
whether the unit is subcritical or supercritical in technology.

It is also interesting to note that the level of availability,
which minimizes capital costs per Kw for a given sized unit, is
frequently lower than the level of availability which minimizes capital
costs per KHH for the unit. In fact, the level of availability that
minimizes capital costs per KWH for a given sized unit is always
greater-than or equal-to the level of availability that minimizes
capital costs per KH for the same unit. Thus, minimizing capital costs
per KH does not assure minimum capital costs per KWH for a given sized

unit.

103

.::x goo mucmu cH moo mmgoowo HH< "muoz

mum mPDMP .m Lmunmnu cw “own—Loam.» 92.:me :memmkmwk cc Ummmm .4.

 

 

 

 

goo.o mmo.o goo." ooo.o ooo.o omo._ oom." mom. moo. mom. mom. gmm
omo. ooo. ooo. moo. mam. oom. omo. oom. mom. moo. mmo. oom
oom. _oo. moo. ooo. mom. oom. omm. mom. mmo. ooo. ooo. goo
oom. oom. omm. oom. omm. omm. oom. “on. “on. mom. omm. goo
~mm. omm. oom. can. mam. omm. oom. mom. «on. mgo. omo. goo
~_m. mom. omm. oom. mom. oom. Nam. ooo. ooo. oom. o_o. goo
oom. omm. mom. «on. "an. ooo. omo. mom. ogm. moo. omo. «mg
"on. com. mom. ooo. ooo. omo. oom. moo. mmo. Hmo. om_.~ goo

.H..>¢ .>.=cu
8o 9: so so So 8... as So So an 8...

mhhosoauz
.:§.8::83.o£=§.8u80::%u.Tozfip

 

104

.zx Loo mgmHHoo cw mom monsoom HH<

”maoz

oYo mHoah .o gmaomco cH omucooms moHommg copmmmgowg co oommo «

 

 

 

 

oo.ooom oo.oooo mm.oooo oo.omoH oo.oooH oH.oom oo.ooo oo.ooo oo.omo oo.ooo oo.HmH gom
oo.ooo oo.ooo oo.oHo oo.ooo oo.oHo mo.ogo oo.ooo oo.ooo oH.ooH mo.ooH oo.moH gom
oo.ooo om.ooo oo.ooo oo.ooo oo.omH om.moH oo.ooH oo.ooH oo.ooH oo.ooH oo.ooH goo
oo.ooH oo.HoH oo.ooH Ho.HoH oo.ooH mm.ooH oo.HoH oo.moH Ho.moH om.HoH oo.ooH goo
oo.ooH oo.ooH HH.ooH oo.ooH oo.ooH oo.ooH oo.HoH oH.ooH oo.ooH oo.ooH Hm.omH goo
oo.moH Hm.HoH mm.omH oo.moH oo.ooH oo.HoH oo.ooH oo.ooH oo.moH oo.oHo oo.moo goo
oo.ooH oo.ooH oo.ooH oo.moH Hm.ooH oo.moH mo.ooH oH.omH oo.mHo oo.ooo oo.oHo goo
oo.mHH oo.ooH oo.omH mo.moH oH.ooH oo.HoH oo.omH oH.mHo om.ooo oH.oHo oH.omo goo
.HHm>< .>H=ou
ooo ooo ooo ooo ooo ooo ooo ooo ooo ooo ooo
ohh<z<oox
y nos: Hmuouogoooo a you xx coo umoo HouHooo .oYo oHoah

 

.zzg Loo mocmu :_ mom mmgoooo HH< "muoz
oYm oHoap .o cmuoaco cH omugoomg maHommg coommmsomg co owmmo «

 

105

omo.o ooo.o ooo.o ooo.o ooo.o ooo.o ooo.H ooH.H mHo. ooo. ooo. Rom
ooo.H ooH.H ooo.H omo. mos. ooo. ooo. ooo. ooo. oHo. Hoo. gom
ooo. Hoo. ooo. oHo. ooo. ooo. ooo. ooo. ooo. “Ho. omo. goo
ooo. omo. Hoo. moo. omo. ooo. Hoo. ooo. moo. oom. ooo. goo
Hoo. Hoo. mom. ooo. Hoo. ooo. mmo. oHo. ooo. ooo. oHo. goo
ooo. oom. Hoo. oom. omo. oHo. ooo. ooo. omo. ooo. oHo. goo
ooo. ooo. ooo. omo. ooo. moo. HHo. ooo. ooo. moo. ooo. goo
Hoo. ooo. omo. ooo. ooo. omo. ooo. Hoo. goo. ooo.H ooo.H goo

.H..>o .>_=oo

 

 

ooo ooo ooo ooo ooo ooo ooo ooo ooo com com
ohh<3<ums

 

grgpzztsaaoezfioaasuztfiu.9o23»

 

106

.:x goo ogmHHoo co moo omgomoo HH<

”moo:

oYo mHoap .o Lwooooo c. omasoomg ouHoomL cooommgomo co ommoo «

 

oo.ooo

 

 

 

oo.omgo oo.oogo oo.momo mo.oooo oo.oogo oo.ogoo oo.oom oo.ooo oo.omo oo.ooo gom
oo.ooo oo.ooo oo.ooo oo.ooo oo.ooo oo.ooo oo.omo mo.moo oo.ooo om.ooo oo.ooo gom
oo.ooo mo.omo oo.ooo oo.ooo oo.ooo oo.ooo oo.ooo oo.ooo oo.ooo oo.ooo oo.ooo goo
oo.moo oo.ooo oo.omo om.omo mo.ooo oo.ooo oo.ooo oo.ooo oo.ooo mm.ogo oo.ooo goo
oo.ooo oo.ooo oo.ooo oo.ooo oo.ooo oo.ooo oo.ooo oo.ooo oo.omo oo.ooo oo.ooo goo
mm.ooo oo.ooo oo.ooo oo.ooo oo.ooo oo.moo mo.ooo om.omo om.go~ oo.ooo oo._m~ goo
oo.ooo oo.ooo oo.ooo mo.moo mo.ooo oo.ooo oo.omH oo.ooo om.oo~ oo.mmo oo.ooo goo
om.mo_ mo.goo oo.ooo oo.ooo oo.oo_ oo.ooo oo.ooo oo.ooo oo.omo oo.ooo oo.ooo goo
._.o>< .ooomo
o8 ooo ooo ooo ooo ooo ooo ooo ooo oom oom
opposoooz
. ooo: Hoo.ooooooo o ooo xx ooo omoo .oooooo .oYo ooooo

 

.132 Loo magma co moo omoomoo HH< "maoz
“Yo mHoap .o gmuoogo co oougooms ouHoomo cooommgmms co omomo «

 

107

ooo.oo ooo.oo omH.Hm omm.oo ooo.oH Hoo.o ooo.o oHo.o gmo.H ooo. moo. gom
ooo.o ooo.o oHo.o ooo.o ooo.H ooo.H omm. ooo. “so. ooo. ooH. gom
ooo.H Hoo.H ooo.H on. ooo. ooo. omo. ooo. mom. ooo. oHo. goo
ooo. ooo. ooo. moo. omo. ooo. moo. mom. ooo. mmo. Hoo. goo
ooo. moo. ooo. ooo. ooo. ooo. ooo. oom. ooo. Hoo. oom. goo
ooo. moo. oom. mmo. ooo. mmo. ooo. oom. oom. ooo. ooo. goo
omo. omo. Hoo. ooo. oHo. ooo. oom. ooo. ooo. mmo. oHo. goo
ooo. ooo. ooo. omo. oHo. ooo. goo. omo. mHo. ooo. oom. goo

._.o>< .ooooo

 

oooH oooH ooHH oooH oom ooo ooo ooo ooo ooo ooo
ophtxtam:

 

 

.. 2.5 33:89:... o .3.— =3. 3.. 38 .328 .oYo as:

 

108

 

 

 

 

.32 Lao moapHoo co woo omgommo HH< “moo:
oYo «Hoop .o gwuomgo co oungoomg ouHoomg economommg cc omooo «
oo.ooooo oH.oomoo oo.ooogH oo.oooHH oH.ogoo oo.oogo oo.oooo om.mooH oo.omo oH.oom oH.omH gom
Hm.oooo oo.mmmH Ho.oooH oH.mooH oo.oom om.oHo oH.Hoo oo.ooo oo.ooo oo.HoH oo.om gom
Ho.oHo oH.Hoo oo.ooo oo.ooo om.ooo oo.ooo mm.ooo om.oHo mo.ooH oo.ooH Ho.ooH goo
oo.Hoo oo.HHo oo.ooo oo.ooo oo.Hoo oo.moo oo.ooH oo.moH oH.ooH oo.moH oH.HoH goo
Ho.ooo oo.omH om.ooH mo.ooH mo.moH oo.ooH oo.ooH oo.ooH oo.ooH oo.ooH oo.ooH goo
om.ooH oo.ooH Ho.ooH oo.omH oH.ooH oo.ooH oo.HoH oo.ooH mo.ooH om.ooH oo.oHo goo
oo.HHH oo.oHH oH.oHH oo.oHH oo.ooH oo.ooH oo.omH Ho.ooH oH.ooH Ho.ooo oo.ooo goo
oo.Hm oo.om om.om oo.ooH oo.moH oH.mHH oo.ooH mo.HoH mm.HoH oo.ooo oH.oom goo
.HHo>< .>.:om

oooH oooH ooHH oooH oom ooo ooo ooo ooo ooo ooo

ohh<z<oox
a ooo: Hoo.».ousomoo o sou 3x goo umoo Huumooo .oYo oHonh

 

109

.::x ooo oucmu co moo owoomoo HH< "moo:
oYo «Hoop .o gmuoago c. omugoomo ouHoomg cooommoomg co omooo «

 

 

 

Hmm.o Hoo.o ooo.o omo.o ooo.o omm.H omo.H omo.H ooo. ooo. Hoo. Rom
ooH.H oHo.H Hmo. ooo. ooo. moo. omo. Hoo. ooo. omo. ooo. gom
ooo. ooo. omo. ooo. ooo. omm. oom. mom. oHo. omo. omo. goo
oHo. omo. oom. ooo. Hoo. ooo. oom. Hoo. oom. ooo. ooo. goo
ooo. mom. oom. moo. Hoo. ooo. oom. ooo. omo. ooo. omo. goo
omo. omo. ooo. ooo. ooo. oom. ooo. ooo. ooo. ooo. moo. goo
ooo. oom. oom. ooo. ooo. mmo. ooo. ooo. Hoo. moo. ooo. goo
ooo. ooo. oom. oHo. ooo. ooo. Hoo. moo. moo. oom. omH.H goo
.—«o>< .>.=um
can com com ooo ooo com com ooo ooo ooo ooo
mhh<3<wux

 

c u—c: —ou.uosoo:m o Lo; :32 Log umou Houoouu .oYo u—oah

 

110

.:x Loo oooHHoo co moo omgomoo HH<

"muoz

oYm mHoop .m omuoogo co omugooms ooHoomg comoomgmmo co omooo «

 

 

 

 

oo.oogo oo.ooom oo.oooo oo.oooo oo.oooo oo.oooo oo.ooo oo.ooo oo.ooo oo.mmo oo.ooo gom
oo.ooo mo.ooo mo.ooo oo.ooo oo.ooo oo.ooo mo.ooo oo.ooo oo.ooo oo.ooo om.mo_ gom
oo.ooo oo.ooo oo.ooo oo.ooo oo.ooo oo.omH oo.ooo oo.ooo oo.ooo oo.ooo oo.ooo goo
oo.omo oo.ooo oo.moH om.ogo oo.ooo mo.ooH oo.ooo mo.moo oo.ooo oo.ooo oo.ooo goo
oo.ooo oo.ooo oo.ooo oo.ooo oo.ooo om.ooo oo.moo oo.ooo oo.ooo oo.ooo oo.ooo goo
mo.oo_ oo.ooo oo.ooo oo.ooo oo.ooo oo.ooo oo.oo_ oo.moH oo.omo oo.ooo oo.ooo goo
oo.ooo oo.ooo oo.ooo mm.ooo oo.ooo oo.ooo oo.mgo om.mmo oo.ooo oo.ooo oo.ooo goo
oo.ooo oo.ooo oo.ooo oo.ooo om.ooo oo.ooo oo.omo oo.ooo oo.ooo oo.ooo oo.ooo goo
...o>< .ooooo
ooo ooo ooo ooo ooo ooo ooo ooo ooo ooo ooo
otozooox
. ooo: .ouoooooooo o ooo 3g ooo omoo .oooooo .oYo o_ooo

 

.::x goo oucmu co moo omgomoo HH< "woo:
oYo mHooh .o gmuoogo co omugoomo ouHoomg coooomgmmg co ooooo «

 

lll

oom.oo moo.oo ooo.nH ooo.oH ooo.o omo.o ooo.o ooH.o moo.H ooo. omo. gom
oom.o ooo.o mom.H omo.H ooo.H omm. goo. mmo. ooo. ooo. oHo. gom
ooo.H ooo. ooo. moo. omo. oHo. mmo. oom. moo. mmo. ooo. goo
ooo. oHo. Hoo. ooo. ooo. ooo. ooo. ooo. HHo. ooo. ooo. goo
ooo. ooo. ooo. mom. moo. ooo. oHo. Hoo. ooo. oom. omo. goo
moo. mmo. omo. omo. mmo. oom. oHo. ooo. ooo. goo. omo. goo
ooo. ooo. Hoo. moo. Hmo. ooo. omo. ooo. ooo. ooo. oHo. goo
ooo. ooo. Hoo. ooo. omo. ooo. ooo. ooo. moo. oHo. ooH.H goo

.H_o>< .>H:uu

 

 

oooH oooH ooHH oooH oom ooo ooh ooo ooo ooo oom
mhh<z<uuz

 

.. :5 33:22.3. o 3o .3. ooo :8 .538 .mYo as:

 

112

 

 

 

 

.zx Loo oooHHoo co ooo omgomoo HH< "mac:
oYo «Hook .o gmuoogo a. omaooomg muHomms coooomomms co umooo «
oo.ooooH oo.ooooH oo.Hmom oo.oooo Ho.oooo oo.HoHo mo.ogmH oo.omHH om.Hoo oH.oom oo.ooH gom
oo.oooH oo.ogoH oo.oooH om.ooo oo.ooo oo.ooo oH.ooo oo.ooo Ho.ooo oo.ooH Hm.HHH gom
oo.mmo Ho.ooo oo.ooo om.ooo oo.omo oo.ooo oo.oHo oo.ooH oo.ooH oo.ooH Ho.ooH goo
om.ooo Ho.ooo mo.mHo oo.ooo oo.ooH oo.ooH om.HoH oH.ooH oo.ooH Ho.ooH oo.moH goo
oo.ooH oo.ooH oo.ooH Ho.ooH oo.ooH oo.ooH oo.moH oo.ooH om.ooH oo.moH mm.moH goo
oo.moH Ho.ooH oo.HoH mH.HoH oo.ooH om.ooH mm.moH oo.ooH oo.ooH oo.ooH Ho.ooo goo
oo.mm oo.ooH oo.ooH oo.ooH oo.oHH mo.oHH oo.ooH oo.ooH oo.ooH oo.oHo oo.oHo goo
oo.oo oo.go oo.Hm oo.om oo.ooH oo.oHH oo.ooH oo.moH oo.ooH oo.ooo oo.omo goo
.Hoooo .ooooo

oooH oooH ooHH oooH oom ooo ooo ooo ooo ooo ooo

ohb<3<ooz
c ooo: Hoo.».susoooo a go» :2 ooo umoo Houoooo .oHYo «Hoop

 

113

.::g Lao oucmu c. moo omoomoo HP< "wuoz
mYo mHooh .o omuooso co omuooomo ouHoomo cooooomomoo co omooo «

 

 

 

 

ooo.o ooo.o oom.o ooo.o oom.o ooo.o ooo.o oom. ooo. ooo. ooo. gom
moo. ooo. ooo. omo. ooo. ooo. ooo. ooo. moo. ooo. ooo. gom
ooo. ooo. omo. ooo. moo. ooo. ooo. ooo. omo. moo. ooo. goo
ooo. ooo. ooo. moo. ooo. ooo. ooo. ooo. ooo. ooo. ooo. goo
omo. omo. ooo. ooo. moo. ooo. ooo. moo. moo. ooo. ooo. goo
ooo. omo. ooo. ooo. ooo. ooo. omo. ooo. ooo. ooo. mmo. goo
omo. ooo. ooo. moo. ooo. ooo. ooo. ooo. ooo. ooo. ooo. goo
ooo. ooo. ooo. ooo. ooo. omo. moo. moo. ooo. omo. ooo.o goo
.ooooo .>.ooo
ooo ooo ooo ooo ooo ooo ooo ooo 8o ooo ooo
oooozoooz
. ooo: ooo.o.ooooo o ooo =32 ooo oooo .oooooo .oHYo ooooo

 

114

 

 

 

 

.zx ooo oooHHoo co moo oooomoo HH< "muoz
mYo oHooo .o cmuoooo co omoooomo ouHoomo oo.oomommo co omooo «

om.oooo oo.oomH Ho.mooH oo.HooH oo.oooH mo.ooo oo.Hmo oo.ooo oo.ooo oo.omo oo.ooo gom
mo.oHo oo.moo oo.ooo oo.oHo oo.ooo oo.ooo oo.ooo oo.omH oH.ooH Ho.moH Ho.ooH gom
om.HHo oo.ooo oo.omH oo.ooH om.ooH mo.moH oo.HoH oo.ooH oH.ooH oo.ooH oo.HoH goo
oo.ooH oo.HoH oo.moH om.ooH oo.ooH oo.ooH om.ooH Ho.ooH oo.moH oo.ooH oo.ooH goo
oH.ooH Ho.ooH om.ooH oo.ooH Ho.moH oo.ooH oo.ooH oo.ooH oo.ooH om.ooH oo.omH goo
oo.oHH om.HoH oo.ooH oo.ooH oo.ooH oo.moH oH.ooH mo.ooH oo.omH oo.oHo oo.ooo goo
oo.oHH oo.mHH oo.ooH oo.ooH oo.ooH oo.moH oo.ooH oo.omH oo.ooo oo.ooo oo.moo goo
oo.oHH oH.HoH om.ooH oo.ooH oo.ooH HH.ooH om.omH oo.ooo oo.Hoo oo.oHo oo.moo goo

.H_o>< .>o:uo
ooo ooo ooo ooo ooo ooo ooo ooo ooo ooo ooo

oph<z<ooz
c coco Honouoouooo a ooo :2 ooo «ooo Houoooo .oHYo oHoop

 

.zzx ooo oocmu co moo omoomoo HH< “mooz
mYo mono» .o owuooso co omaooomo ooHoooo ooooomommo co omooo «

 

115

ooo.oo ooo.oo omo.oo ooo.oH moo.oH oHo.m Hoo.o oom.m ooo.o oHo.H ooo. gom

ooo.o oom.o ooo.o ooo.o omo.H ooo.H ooo.H oHo. omo. oHo. ooo. gom
oHH.H ooo.H oom. ooo. oHo. moo. omo. moo. oom. oHo. ooo. goo
omo. ooo. Hoo. ooo. ooo. moo. Hmo. oom. Hoo. oom. ooo. goo
omm. oom. oom. ooo. oom. ooo. omo. Hoo. ooo. ooo. omo. goo
oHo. moo. ooo. moo. HHo. oHo. Hoo. Hmo. ooo. ooo. mHo. goo
ooo. moo. ooo. ooo. omo. ooo. oom. mom. ooo. moo. omo. goo
ooo. ooo. Hoo. ooo. omo. oHo. mom. oom. moo. moo. omo. goo

.H.o>< .>_=oo

 

 

ooo" ooo" cog" oooH oom ooo ooo ooo ooo ooo ooo
oph<3<aux

 

.. :5 .832233 o ooo =3. ooo 38 .338 .24 as:

 

116

 

 

 

 

.zx Loo ooopHoo co moo omooooo HF< "muoz
mYo mpooh .o omuoogo co omoooomo oopoooo coooomommo co omooo «
oo.ooooo oo.ooooH oo.ooHoH oo.ooooH om.oHoo oo.oooo oo.oooo Hm.ooHo oo.oooH oo.ooo mo.ooo gom
oo.oHoH oo.oooH oo.oooH om.oooH mo.moo oo.oHo om.ooo oo.ooo oo.oHo oo.oHo Ho.moH gom
om.ooo oo.ooo mo.moo oH.ooo oo.ooo oo.moo mo.ooo Ho.ooo oo.omH oH.ooH mo.ooH goo
oo.ooo oo.moo oo.ooo Ho.ooo oo.oHo oo.omH Ho.ooH mo.ooH oo.ooH oo.HoH Ho.moH goo
oo.ooH oo.moH oo.ooH om.moH oo.ooH oo.HoH oo.ooH oo.ooH om.ooH oo.ooH oo.ooH goo
oo.ooH oH.ooH HH.ooH oo.ooH oo.ooH oo.ooH oo.HoH oo.ooH oo.HoH HH.ooH oo.HoH goo
oo.ooH oH.ooH mo.ooH oo.ooH Ho.HHH oo.oHH oo.ooH oo.ooH oo.ooH mo.ooH om.ooo goo
oo.oo oo.oo oo.Hm mo.om mo.ooH om.moH oo.ooH oo.ooH oo.ooH oo.ooH oo.Hoo goo
.Hoo>< .>o:uo

oooH oooH ooHH oooH oom ooo ooo ooo ooo ooo ooo

ohh<3<oox
« uHoo Hoo.».ouooooo a 50$ :2 ooo unoo Houoooo .oHYo oHooh

 

117

Another general result of our research is that capital costs per Kw
for both generation technologies fall continuously as unit size
increases, but only at low levels of equivalent availability (GO-70%).
At high levels of equivalent availability (85-95%), capital costs per Kw
increase with unit capacity. Capital costs per KN follow a U-shaped
pattern as unit size increases at intermediate levels of availability
(75-80%). These relationships are clearly seen in Figure 4-1, which
plots capital cost per KU as unit size increases while equivalent
availability is maintained at levels of 60, 75, and 90 percent. Capital
costs per KHH follow very similar patterns for the various combinations
of unit size and availability (see Figure 4-2). These results are in
sharp contrast to those obtained by previous researchers, who generally
found that capital costs per KH fall as unit size increases; however,
they failed to treat reliability as an attribute of a generating unit.

To the extent that the goal of regulators is to minimize capital
costs per KHH, there are two broad groups of size-reliability
combinations that appear to generally satisfy this goal. One group
consists of units in the 300 to 500 NH range with equivalent
availabilities of 80 to 90 percent. The second group consists of very
large units (800 NH and larger) with equivalent availabilities ranging
from 60 to 70 percent. It is interesting to note that capital costs per
KHH are almost always lowest for units in the 300-400 M" range with
equivalent availabilities of 85 to 90 percent.

-——OU

ll

'10

279

118

Dollars/KW

 

 

 

60 Percent

 

 

 

 

 

 

 

 

 

 

90 Percent
600
A
I
I
I
’I
’I
I
400 2‘
I
’I
’I
’I
’l
l”
I”
200 ..... ,7
..";"-€.f ............
y”’ nnnnnnnnnnn
0 T T T T T T T T j
300 350 400 450 500 600 650 700 750 800
MW

Figure 4.1 Capital Cost Per KW

'0 730-9300

Iéz

119

Cents/KWH

 

 

60 Percent

90 Percent

 

 

 

1.5

 

 

 

 

 

 

 

 

 

0 l 1 r l I T l T

300 350 400 450 500 600 650 700
MW
Figure 4.2 Capital Cost Per KWH

750

800

120
or . u' . ' 1 °‘ 'lu-l : -l‘ l‘ i ~-; I -b i .deof

Poor reliability means that the capital-related costs of a baseload
generating unit are spread over a lower level of output than if the unit
is used more intensively. Poor reliability also reduces the thermal
efficiency of a generating unit in two ways. First, frequent deratings
of a unit due to outages means that heat energy must be expended to
reheat the boiler and other components. Second, heat loss is relatively
larger at low load than at high load, given that the absolute amount of
heat loss is fairly constant. Thus, poor unit reliability means higher
average costs per KHH generated.

As a result, a number of state utility commissions have initiated
incentive programs aimed at improving the equivalent availability of a

° The idea is to encourage

utility’s baseload generation facilities.7
the utility to keep a unit running as much as is economically
reasonable. One approach is to tie a utility’s return on equity to the
level of plant availability. For example, a normal range of plant
availability may be set between 70 and 80 percent. Performance below 70
percent causes the return on equity to be reduced by .25%, while
performance greater that 80 percent is rewarded by allowing the return
on equity to increase by .25%.

However, setting minimum standards for a narrow range of
performance criteria can create perverse incentives for a utility."' A
utility might be willing to incur excessive costs in areas outside the

incentive program so as to improve performance in the targeted areas.

For example, a utility might spend excessively on maintenance and, thus,

121
partially or even totally offset the benefits to rate payers of higher
unit availability.

The results given above and in Chapter 3 reveal that there is a
relationship (or connection), everything else equal, between the
construction costs and reliability of a generating unit. These results
indicate that there is an optimum level of reliability depending on
generation technology and unit-size which minimizes capital costs per
KHH and/or capital costs per KN. An examination of Tables 4—1 through
4-14 also reveals that capital costs can increase significantly if the
desired level of equivalent availability is either higher or lower than
the cost-minimizing level of reliability for a given sized unit. Thus,
to the extent that a utility has an incentive to increase the
reliability of a unit under construction beyond the cost-minimizing
level for a given unit-size due to an incentive program, the additional
capital costs may more than offset any potential benefits to ratepayers.
As a result, state regulators should also develop policies that focus on
unit design and the construction process.

Traditionally, however, state utility commissions have relied on
prudence tests to help offset the disincentive effects of cost-plus
regulation on utility capital spending. The disincentive problem is
accentuated by the fact that regulators generally have less information
than utility management regarding utility investment decisions and the
efficiency of the generating unit design and construction process. The
prudence test is an imperfect tool to improve utility performance,
because it can only be used to punish especially bad and costly

outcomes.

122

Thus, the FERC has proposed all-source bidding programs for new
generation capacity in an effort to improve the efficiency of utility
investment and construction decisions]" The idea is that a competitive
solicitation for new generation capacity from all sources will promote
the construction of the least cost facilities.

The general nature of an all-source bidding program that may result
from the FERC proposal can be seen from the broad characteristics of a
number of state bidding programs that currently existJ" Under
all-source competitive bidding, a utility would forecast its need for
additional generation capacity and set the long-run avoided cost cap for
the new electricity. The cost cap would be set equal to the utility’s
projected cost of supplying the additional electricity itself. At this
point, the utility would request proposals for generation capacity from
other sources. If alternative generation sources are inadequate to meet
the utility’s needs or the utility’s own offer is deemed best in terms
of cost and reliability, then the utility would be permitted to build
the needed generation facilities. If the utility spends more than the
avoided cost cap, only the amount stated in the cap would be added to
the rate base. Should the utility spend less than the cap, it would
earn a higher rate of return, since the total cap amount would be
included in the rate base.

However, at least two issues must be settled before any type of
bidding program can be implemented. First, to what extent should
generating capacity choices be made on the basis of price? Second, to
the extent that noncost factors are included in the bid evaluation

process, what are the appropriate noneconomic factors to be included and

123
how are they to be evaluated?

Current state bidding programs generally utilize some type of
ranking system to evaluate and compare bids. A common practice is to
divide the ranking criteria into two broad categories--economic and
noneconomic. The economic factors usually receive the greatest weight
in the ranking process and are primarily limited to the bid price of the
electricity. Noneconomic factors typically include the following: (1)
project schedule and milestones, (2) project financing, (3) project team
and experience, (4) fuel type, (5) generation technology, (6)
engineering design, and (7) reliability.“’

It is important to obtain a balance between price and nonprice
considerations and to conduct a thorough evaluation of a bid to
determine whether the proposed price of electricity is consistent with
the noneconomic factors of the bid. This is necessary because the
“noneconomic” factors of a proposed generating facility can have a
significant impact on the “economic” factors. For example, the
econometric analysis in Chapter 3 highlighted the simultaneous nature of
the relationship between the reliability (a common noncost factor) and
construction costs of a generating unit. But the ability to make simple
and straightforward checks on the consistency between the economic and
noneconomic factors of a bid and to compare the characteristics of
alternative bids remains to be seen.

One way that utilities are coping with this uncertainty is by
inserting a number of conditions into contracts with independent bidders
for new generation capacity. These conditions are aimed at reducing the

uncertainty that utilities have regarding whether independent suppliers

124
will prove to be reliable sources of generation capacity. Among the
conditions being contractually required by some utilities are: (1) large
security deposits, (2) warranties from engineers and equipment
manufacturers, (3) site inspections, (4) financial audits, and (5)
supervision of maintenance."

State utility commissions and purchasing utilities are also
requiring independent suppliers to satisfy performance guarantees in
order to avoid paying penalties. A fairly typical standard requires the
owner of an independent generation facility to pay a capacity penalty if
the facility fails to maintain an annual average availability factor
greater-than or equal-to the lesser of the purchasing utility’s prior
year’s weighted average of equivalent availability for its non-nuclear
units or a cap of 80 percent for solid fuel facilities.82

However, to the extent that state regulators seek high levels of
unit reliability (defined as 80% and higher) and reasonable capital
costs, it would seem desirable to encourage the construction of baseload
coal-fired units in the 300-450 HH range. An examination of the above
tables reveals that both capital costs per KHH and capital costs per KH
increase as unit size increases beyond 450 MH while maintaining an 80%
level of equivalent availability. At levels of equivalent availability
greater than 80%, both measures of capital costs increase as unit size

increases beyond 300 NH.

§HEMQLI
In this chapter we have reviewed how capital costs per KH and

capital costs per KHH respond to variations in unit size and

125
reliability. A general result is that the level of reliability which
minimizes capital costs per KHH is always greater than or equal to the
level of reliability which minimizes capital costs per KH for a given
sized unit. Another general result is that the level of reliability
which minimizes each measure of capital costs falls as unit size
increases. Finally, both measures of capital costs either fall, rise,
or follow a U-shaped pattern as unit size increases beyond 300 HH while
maintaining, respectively, low (GO-70%), high (85-95%), or intermediate
(75-801) levels of equivalent availability.

He also used these results to briefly review and evaluate two
regulatory reforms aimed at improving incentives for efficiency in the
electric utility industry. Maintaining a high level of unit reliability
was the primary objective of one program and an important objective of
the other program. Each program sought to promote high levels of unit
reliability by tying some combination of financial rewards and/or
penalties to the average annual equivalent availabilities of specified
units.

Our results indicate, however, that if state and federal regulatory
agencies desire high levels of unit reliability and reasonable capital
costs, then they should promote the construction of coal-fired baseload
units in the 300-450 MH range with equivalent availabilities of 80 to 90
percent. In fact, capital costs per KHH generated are almost always
lowest for units in the 300-400 HH range with equivalent availabilities
of 85-90 percent.

CHAPTER 5

CONCLUSIONS

§HEM§L¥

The primary objective of this study was to develop a construction
cost function for coal-fired steam-electric generating units which
explicitly recognized that unit size and reliability are inversely
related. The basic concept was that the construction cost of a
generating unit, or any other type of capital equipment, should depend
on the various engineering attributes of the facility. Most previous
studies have treated generating units as if they were homogeneous pieces
of capital equipment that could be represented by a scalar aggregate
measure like unit size. Other studies have generally disaggregated the
engineering attributes of generating units along the dimensions of size
and steam pressure conditions (or heat rate). But the failure to
include reliability as an engineering attribute has meant that it is
impossible to determine whether large units cost less to build because
of economies of scale or because of poorer quality.

To begin our analysis, we noted that the cyclical demand for
electricity means that generating units will be designed for different
modes of operation. Baseload units are designed for high thermal
efficiency and continuous operation at maximum capacity whenever they
are available. At the other extreme, peakers are designed for rapid
starts and shutdowns, lower thermal efficiency, and operation only

during periods of peak demand. A utility’s decision as to what portion

126

127
of the total demand for electricity the unit would serve was seen as
having a significant effect on the desired engineering characteristics
of the unit. Thus, we assumed for the remainder of our analysis that we
were dealing exclusively with coal-fired, baseload units.

Another important feature of our analysis was an examination of how
the unit design process and, thus, the choice of engineering
characteristics shaped the ex post production relationship. Both the
labor-output and fuel-output relationships were found to be conditional
upon the design characteristics of the generating unit. Once a unit is
built, there is very little opportunity for substitution among the
inputs so the technology for steam-electric generation was seen to be of
a putty-clay nature.

As a result, a utility was viewed as being faced with the task, at
the blueprint stage, of choosing unit attributes that minimize the
expected total costs of serving a baseload demand for electricity. The
capital intensive nature of baseload, coal-fired generation means that
the average total generation costs of a unit can be reduced
significantly if it is used intensively. However, baseload units are
meant to be operated at maximum capacity whenever they are available, so
the primary determinant of the intensity with which a baseload unit is
utilized and, thus, an important engineering attribute is the
reliability of the unit.

On the basis of these observations, we developed an ex ante
long-run unit reliability and construction cost model. Both unit
reliability and construction costs were expressed as functions of the

dominant design characteristics of the generating unit. We further

128
noted that a utility can be expected to modify its choice of unit
attributes if it has some conception of the magnitude and sign of the
disturbance term in the construction cost function. Such an event was
deemed reasonable given the substantial experience major utilities have
building and operating a variety of generating units. So a simultaneous
relationship between construction costs and reliability was assumed.

The simultaneous-equation model was estimated with a data set that
contained observations on 84 coal-fired units that entered commercial
operation during the period 1964-1974. He then used regression
estimates to evaluate how capital costs per KH and capital costs per KHH
responded to variations in unit size and reliability.

One interesting result was that units that entered commercial
operation later in the observed period were more reliable than similar
units that entered commercial operation earlier. He also found that
unit size and reliability are inversely related.

Both measures of capital costs were found to be characterized by
economies of scale at low levels of reliability and diseconomies at high
levels of reliability. Both capital cost measures followed a U-shaped
pattern as unit size increased at intermediate levels of reliability.
The cost-minimizing level of reliability for each measure was also found
to decrease as unit size increased. These results indicate that high
levels of reliability are expensive for large units to attain, given the
general size-reliability tradeoff noted earlier.

He also learned that focusing on minimizing capital costs per KH is
inappropriate if utilities and regulators desire to minimize average

total generation costs for a baseload unit. The level of reliability

129
that minimizes capital costs per KHH was found to be greater-than or
equal-to the level of reliability that minimizes capital costs per KH
for a given sized unit. The reason for this result is, the relative
increase in construction costs caused by the higher level of reliability
is less-than or equal-to the relative increase in potential KHH
generated.

Capital costs per KHH were found to be minimized by units that fell
into two broad-size reliability groups. One group consisted of units in
the 300-500 MH range with equivalent availabilities of 80-90 percent.
The second group consisted of units 800 HH and larger with equivalent
availabilities of only 60-70 percent.

It is important to remember that we limited our analysis to the
impact the size—reliability tradeoff has on the average generation costs
of an individual baseload unit. Any consideration given to building
extremely large units with low levels of reliability would be put into a
less favorable light if we were to expand our analysis to include the
impact on system-wide costs and reliability. The lower average
reliability of larger units adds to a utility’s total system costs,
because additional capacity must be built if a specified level of system
reliability is to be maintained.

Perhaps the most important conclusion drawn from this study is that
the size-reliability tradeoff, which seems to characterize the
coal-fired baseload generation technologies, cannot be ignored if
regulatory authorities desire both high levels of reliability and
reasonable capital costs. Current policies promoting improved

reliability by tying financial rewards and/or penalities to the average

130
annual equivalent availabilities of specified units are inadequate,
because they create perverse incentives for a utility. In general,
regulatory agencies need to develop policies that focus on unit design
and the construction process. Regulatory authorities, to be more
specific, should promote the construction of coal-fired baseload units

in the 300-450 HH range with equivalent availabilities of 80-90 percent.

MW

There were two significant limitations placed on the scope of our
analysis which provide interesting areas for future research. First, we
excluded from our analysis nuclear units. Nuclear units are definitely
baseload facilities and have low fuel costs but very high capital costs.
In fact, nuclear units are considerably more capital-cost intensive than
coal-fired units. Thus, the existence of any size-reliability tradeoff
for nuclear units can have a significant impact on their average total
generation costs.

The second area of future research was touched on earlier. Our
primary concern in this study was an analysis of how the
size-reliability tradeoff affected construction cost economies at the
level of the individual unit. However, one of the primary determinants
of system reliability is the reliability of the systems’ individual
generating units. The more frequent a generating unit is broken down,
everything else equal, the lower the reliability of the utility system
and/or the higher are system-wide production costs. Thus, an

interesting area of future research involves analyzing how movement to

131
relatively small (300-450 HH), reliable, baseload, coal-fired units
affects system-wide reliability and costs.

APPENDICES

APPENDIX A

APPENDIX A
THE DEFLATION PROCESS

The following formula was used to deflate the nominal construction

cost of a unit and to net out interest charges:

Construction cost in constant dollars net of interest charges =

___Bg22£tgd_flom1nal Cost
3:, sY- 9:1,“ + pm) 31:1: (1 + rum

 

Hhere,

S,-=the share of actual construction costs in year t (taken from a
typical cash flow curve reported in Power Plaat Capital Costs,
Current Trends and Sensitivity to Economic Parameters, U.S. Atomic
Energy Commission, HASH-1345, October 1974, Figure 5. This gives
annual cash flows for a 5-year construction period.),

P(i) - the percentage change in input prices in year i (taken from

the n -Hh tm bl i i n t tion 0 t ndex),

r(j) - the average allowance for funds used during construction

[rate from the DOE’s Statistics of Privately-Owned utilities in the
United Statas (various years)].

132

APPENDIX B

APPENDIX D

ESTIMATION OF A COBB-DOUGLAS SPECIFICATION
OF THE COST FUNCTION

Below we specify the relationship between average construction cost
and unit size to be Cobb-Douglas, as was done by Joskow and Rose (1985).
He estimate three variations of this basic construction cost
relationship. He initially estimate the cost function using OLS while
excluding the reliability variables and including architect-engineer
dummy variables (see Table B-l). This is similar to the basic cost
function specification estimated by Joskow and Rose. Several of our
results are similar to their findings. He obtain a very large and
precise estimate of the first-unit effect, which implies that such units
are in the range of 25% more costly than follow-on units. He also find
that real costs per KH, net of input price changes, have increased over
time. The coefficient on the time trend variable, ln(year), is positive
and statistically significant at the 1% level. He also found that the
architect-engineer (A-E) dummy variables are jointly significant at the
5% level (F(15,54) - 2.15).

A couple of our results are also different from Joskow and Rose’s
findings. He find that the coefficients of the A-E experience variables
have the expected signs but are not statistically significant. Joskow
and Rose found the experience effects for the supercritical technology
to be fairly large and statistically significant. Finally, we are
unable to reject the null hypothesis that the intercept and scale terms

for subcritical and supercritical units are the same (F(2,54) - 1.23).

133

134

Table 8-1. Ordinary Least Squares Estimate of the
Construction Cost Function

 

 

VARIABLE WW
CONSTANT -10.269 3.139
PRESSURE DUMMY 1.798 1.234
LN(MH) -.138 .140
LN(MH)*PRESSURE -.303 .195
FIRST UNIT DUMMY .275 .055*
LN(YEAR) 3.770 .789*
COOLING TONER DUMMY .070 .061
FULL-INDOOR DESIGN DUMMY .054 .084
A-E SUBCRITICAL UNIT EXPERIENCE -.011 .060
A-E SUPERCRITICAL UNIT EXPERIENCE -.O46 .096
MIDDLE ATLANTIC DUMMY .031 .111
HEST NORTH CENTRAL DUMMY .034 .135
EAST NORTH CENTRAL DUMMY .273 .257
ROCKY MOUNTAIN DUMMY ' - 033 .166
SOUTH ATLANTIC DUMMY - 167 .095-
SOUTHERN SERVICES C0. DUMMY - 285 .121+
TVA CO. DUMMY - 014 .148
DUKE C0. DUMMY -.126 .147
AEP C0. DUMMY .280 .163-
STEARNS & ROGER C0. DUMMY - 108 .226
BECHTEL CO. DUMMY - 133 .138
EBASCO C0. DUMMY - 189 .098-
SARGENT & LUNDY CO. DUMMY - 178 .102-
BROHN & ROOT CO. DUMMY - 106 .292
FLUOR CO. DUMMY - 500 .174*
BLACK & VETCH CO. DUMMY - 296 .168-
GILBERT CO. DUMMY - 170 .133
GC CO. DUMMY - 042 .150
COMMON CO. DUMMY - 145 .138
UNITED CO. DUMMY - 291 .252

 

* - Significant at 1 Percent
+ - Significant at 5 Percent
- - Significant at 10 Percent
FF - .72

Adjusted R2 - .57

F(30, 54) - 4.795

 

135
Joskow and Rose found supercritical units to be characterized by a
higher level of costs and larger estimated scale effects than
subcritical units.

As done in Chapter 3, we now use OLS to estimate the cost function
when our basic reliability variables are added to the cost equation (see
Table 8-2). The first unit effects are once more positive and
significant at the 1% level. Again, the time trend variable has a
positive and significant coefficient (at the 1% level). The size
variable, ln(MH), has a considerably larger coefficient in absolute
value than before and is now significant at the 10% level. Finally, the
reliability variables are not jointly significant (F(3,51) - 1.31).
These results are similar to those reported in Chapter 3.

He will now use 2SLS to estimate the cost function, which includes
the basic reliability variables (see Table B-3).“’ The ZSLS estimate of
the coefficient of -1n(1-EA) is negative and statistically significant
at the 1% level. The coefficient of (-1n(1-EA))2 is positive and
significant at the 5% level. The interaction term, ln(NH)*(-ln(1-EA)),
has a positive coefficient that is significant at the 5% level. It is
important to note that the ZSLS estimate of these coefficients is much
larger in absolute value and has a higher level of statistical
significance than the OLS parameter estimates presented in Table 8—2.

The coefficient on the size variable, ln(MH), has a negative sign
and is significant at the 1% level. The 2SLS estimate of the size
effect is quite larger in absolute value and has a higher level of
significance than the OLS estimate. The time trend variable has the
expected positive coefficient, but it is not significant. The OLS

136

Table 8-2. Ordinary Least Squares Estimate of the
Construction Cost Function

 

 

VARIABLE WW
CONSTANT -3.539 5.554
PRESSURE DUMMY .544 1.597
LN(MH) -1.007 .542-
PRESSURE*LN(NH) -.125 .249
-LN(1-EQUIV. AVAIL.) -3.958 2.450
(-LN(1-EQUIV. AVAIL))2 .220 .135
LN(HH)(-LN(1-EOUIV. AVAIL.)) .520 .344
FIRST UNIT DUMMY .285 .055*
LN(YEAR) 3.618 .872*
COOLING TOHER DUMMY .078 .054
FULL-INDOOR DESIGN DUMMY .029 .085
A-E SUBCRITICAL UNIT EXP. -.024 .055
A-E SUPERCRITICAL UNIT EXP. -.072 .098
MIDDLE ATLANTIC DUMMY .013 .113
HEST NORTH CENTRAL DUMMY —.014 .147
EAST NORTH CENTRAL DUMMY .284 .258
ROCKY MOUNTAIN DUMMY -.051 .159
SOUTH ATLANTIC DUMMY -.155 .095-
SOUTHERN SERVICES C0. DUMMY -.280 .122+
TVA CO. DUMMY .080 .155
DUKE C0. DUMMY -.159 .154
AEP CO. DUMMY .235 .194
STEARN a ROGER C0. DUMMY - 232 .241
BECHTEL CO. DUMMY - 123 .138
EBASCO CO. DUMMY - 193 .103-
SARGENT a LUNDY CO. DUMMY - 151 .103
BROHN a ROOT CO. DUMMY - 205 .301
FLUOR CO. DUMMY - 547 .187*
BLACK 5 VETCH C0. DUMMY - 323 .172-
GILBERT C0. DUMMY - 173 .132
GC CO. DUMMY - 035 .152
COMMON C0. DUMMY - 202 .145
UNITED C0. DUMMY - 317 .250

 

* - Significant at 1 Percent
+ - Significant at 5 Percent
- - Significant at 10 Percent
R’ - .74

Adjusted R2 - .577

F(33, 51) - 4.536

 

137

Table 8-3. Two-Stage Least Squares Estimate of the

Construction Cost Function

 

 

VARIABLE

CONSTANT
PRESSURE DUMMY

LN(NH)

PRESSURE*LN(MH)
-LN(1-E0UIV. AVAIL.)
(-LN(1-EQUIV. AVAIL))2
LN(HH)(-LN(1-EQUIV. AVAIL))
FIRST UNIT DUMMY

LN(YEAR)

COOLING TOHER DUMMY
FULL-INDOOR DESIGN DUMMY
A-E SUBCRITICAL UNIT EXP.
A-E SUPERCRITICAL UNIT EXP.
MIDDLE ATLANTIC DUMMY

HEST NORTH CENTRAL DUMMY
EAST NORTH CENTRAL DUMMY
ROCKY MOUNTAIN DUMMY

SOUTH ATLANTIC DUMMY
SOUTHERN SERVICES C0. DUMMY
TVA C0. DUMMY

DUKE C0. DUMMY

AEP CO. DUMMY

STEARNS 3 ROGERS CO. DUMMY
BECHTEL CO. DUMMY

EBASCO C0. DUMMY

SARGENT a LUNDY CO. DUMMY
BROHN a ROOT CO. DUMMY
FLUOR CO. DUMMY

BLACK 1 VETCH CO. DUMMY
GILBERT CO. DUMMY

GC CO. DUMMY

COMMON C0. DUMMY

UNITED C0. DUMMY

EQEEEIEIENI

.084
.964
.053

ANDA ERROR

12.
3.
1.

.492
5.

090
175
104*

273*

.292+

.793+
.067*
.588
.098
.107
.112
.132
.150
.239
.333
.219
.129
.164+
.202
.273-
.411
.331
.179
.173-
.137
.401
.277+
.245-
.163
.196
.203
.389-

 

* - Significant at 1 Percent
+ - Significant at 5 Percent
- - Significant at 10 Percent
R2 - .672

Adjusted R2 - .467

F(33, 51) - 3.271

 

138
estimate of the time effect is larger in absolute value and has a higher
level of significance.
These results are basically the same as those reported in Chapter
3. They suggest that the potential bias from ignoring the simultaneous
nature of the model and/or ignoring the measurement error associated

with the use Of realized reliability instead of the ex ante level of

 

reliability may be important factors in the estimation of generating

unit construction cost/unit reliability relationships.

139

Table 8-4. Ordinary Least Squares Estimate of the
Construction Cost Function

 

 

VAAlAflLE W M AR” RROR
CONSTANT -19.789 7.550
PRESSURE DUMMY -.091 .315
LN(MH) 2.980 2.225
(LN(MN))2 -.259 .175
FIRST UNIT DUMMY .280 .055*
LN(YEAR) 3.807 .791*
COOLING TOHER DUMMY .055 .052
FULL-INDOOR DESIGN DUMMY .052 .085
A-E SUBCRITICAL UNIT EXPERIENCE -.043 .057
A-E SUPERCRITICAL UNIT EXPERIENCE -.050 .095
MIDDLE ATLANTIC DUMMY .031 .111
HEST NORTH CENTRAL DUMMY .031 .135
EAST NORTH CENTRAL DUMMY .305 .257
ROCKY MOUNTAIN DUMMY -.053 .154
SOUTH ATLANTIC DUMMY -.168 .095-
SOUTHERN SERVICES C0. DUMMY -.289 .121+
TVA CO. DUMMY .021 .157
DUKE CO. DUMMY -.140 .147
AEP C0. DUMMY .353 .174+
STEARNS a ROGER C0. DUMMY - 095 .225
BECHTEL CO. DUMMY - 105 .133
EBASCO c0. DUMMY - 194 .099-
SARGENT a LUNDY CO. DUMMY - 170 .102-
BROHN a ROOT CO. DUMMY - 092 .291
FLUOR C0. DUMMY - 519 .174*
BLACK 1 VETCH C0. DUMMY - 327 .159-
GILBERT C0. DUMMY - 155 .132
CC C0. DUMMY - 053 .150
COMMON C0. DUMMY - 139 .138
UNITED C0. DUMMY - 302 .254

 

* - Significant at 1 Percent
+ - Significant at 5 Percent
- - Significant at 10 Percent
R2 - .719

Adjusted R2 - .568

F(30, 54) - 4.764

 

140

Table 8-5. Ordinary Least Squares Estimate of the

Construction Cost Function

 

 

VARIABLE CREEEIEIENI
CONSTANT -15.535
PRESSURE DUMMY 1.157
LN(MH) 1.555
(LN(MN))2 -.139
PRESSURE*LN(MH) -.201
FIRST UNIT DUMMY .279
LN(YEAR) 3.787
COOLING TONER DUMMY .058
FULL-INDOOR DESIGN DUMMY .050
A-E SUBCRITICAL UNIT EXP. - 022
A-E SUPERCRITICAL UNIT EXP. -.051
MIDDLE ATLANTIC DUMMY .032
NEST NORTH CENTRAL DUMMY .039
EAST NORTH CENTRAL DUMMY .285
ROCKY MOUNTAIN DUMMY -.040
SOUTH ATLANTIC DUMMY -.170
SOUTHERN SERVICES CO. DUMMY -.286
TVA CO. DUMMY .015
DUKE CO. DUMMY -.125
AEP C0. DUMMY .323
STEARN a ROGER C0. DUMMY - 110
BECHTEL CO. DUMMY - 132
EBASCO C0. DUMMY - 197
SARGENT a LUNDY CO. DUMMY - 177
BRONN a ROOT C0. DUMMY - 113
FLUOR CO. DUMMY - 505
BLACK 3 VETCH C0. DUMMY - 311
GILBERT CO. DUMMY -.168
CC CO. DUMMY -.O49
COMMON C0. DUMMY -.145
UNITED C0. DUMMY - 310

 

* - Significant at 1 Percent
+ - Significant at 5 Percent
- - Significant at 10 Percent
R2 - .721

Adjusted R2 - .565

F(30, 53) - 4.59

 

APPENDIX C

APPENDIX C
DERIVATION AND STATISTICAL SIGNIFICANCE OF NONLINEAR DERIVATIVES

If

inPk = 80 + 81ln(MH) + 82(ln(MH)2 + B3(-ln(1-EA) +

84(-ln(1-EA))2 + 85(-ln(1-EA))ln(MH) +

85Pressure + 87(Pressure*ln(MH)) +

88(Pressure*(-ln(1-EA))) + . . .

(Hhere the omitted variables are linear), then

alnPk/alnMH = 81 + 282ln(MH) + 85(—ln(1-EA)) +

87Pressure.

For any particular (ln(MH), -ln(1-EA), Pressure) and estimated (81,
82, 85, 87), we Obtain an estimated alnPk/alnMw, with variance

Var(alnPk/8lnMH) = Var(81) + 4Var(82) (In(MN))2 +

Var(85) (-ln(1-EA))2 + Var(87) (Pressure)2 +

4C0V(Bl,82) (lnMw) + 2Cov(81,85) (-ln(1-EA)) +

2Cov (81,87)Pressure + 4C0v(82,85) (ln(MH)*(-ln(1-EA))) +

4Cov(82,87) (ln(MW)*Pressure) +

2Cov(85,87) (-ln(1-EA)*Pressure).

Var is the variance of 81 and COV is the covariance of 81 and Bj.
These are elements of the parameter variance-covariance matrix. A T-
statistic under the null hypothesis that alnPk/alnMH is then

(alnPk/alnMH)/SE
Hhere SE is the standard error Of the estimated elasticity, or the

square root of the variance.

141

142
Similar computations were done for the elasticity Of unit

construction cost with respect to reliability.

END NOTES

END NOTES
CHAPTER 1. INTRODUCTION

1. The National Electric Reliability Council (NERC) defines
equivalent availability as the fraction of a period that a unit is
available to generate power, adjusting for partial outages that reduce
the effective capacity. The NERC is the source of our equivalent
availability data.

2. Joskow and Rose (1985), Cowing (1974), Komanoff (1981), Perl
(1982), Stewart (1979), and Hills (1978) are more recent studies of the
construction cost Of coal-fired generating units.

3. Joskow and Rose (1985) examined the average size Of new units
placed into commercial Operation between 1950 and 1982. They found that
average unit size increased fairly rapidly until 1975 and fell slightly
over the period 1975-82.

4. This was based on the engineering rule of thumb called the
"six-tenths factor" which states that the capital cost of a new
generating unit increases only in proportion to the six-tenths power of
the capacity of the unit.

5. See Perl (1982), Corio (1982), and Joskow and Schmalensee
(1987).

6. See Komanoff (1976) and EPRI (1982).

7. See Komanoff (1976) for a nontechnical discussion of the
engineering basis for this relationship. Joskow and Schmalensee (1987)
present econometric results which support this relationship.

8. See DOE (1985) and Joskow and Rose (1985) for good discussions
of these design characteristics.

9. See Nagelhaut (1988) for a survey of developments in those
states in which bidding processes have been implemented or are under
consideration.

10. See Johnson (1985) and Joskow and Schmalensee (1986).
CHAPTER 2. SURVEY AND CRITIQUE OF THE LITERATURE

11. For a more detailed discussion of steam-electric generating
technology, see Ling (1964), Cowing (1974), and Bushe (1981, Ch. 2).

12. See Bushe (1981), pp. 63-65.

13. For a more detailed discussion of technological Change and the
innovation process, see Bushe (1981, Ch. 3) and Smith (1977).

14. See Bushe (1981, Ch. 3) and Joskow and Rose (1985) pp. 3-4.

143

144

15. The same criticism was made by Bushe (1981, p. 125).

16. Barzel (1964), p. 142.

17. This problem was noted by Barzel, pp. 142-143.

18. This criticism was also made by Houldsworth (1985).

19. Komanoff (1976) found empirical support for a break- in period
for coal- fired units. But Joskow and Schmalensee (1987) found no
empirical evidence Of a break-in period.

20. Galatin (1968), p. 131.

21. Ibid., pp. 133-134.

22. That the desired level of utilization affects plant capital
costs independent of size was noted by Houldsworth (1985) and Huettner
(1964).

23. Huettner (1974), p. 47.

24. Ibid., p. 47.

25. Ibid., p. 98.

26. Hills (1978), p. 504.

27. The studies in this section generally avoid the vintage
problem by using samples that contain only coal- fired units built Since
1960. Joskow and Rose (1985, pp. 3- -4) note that since 1960 the "primary
technological frontier" has been in the steam pressure dimension. All
coal-fired units built since 1960 fall into four pressure classes
grouped around 1800, 2000, 2400, and 3500 PSI. Thus, an adequate
control for "vintage“ simply requires only a steam pressure variable or
a heat rate variable.

28. Stewart (1979, p. 552).

29. The same criticism was made by Joskow and Schmalensee (1983,
p. 229).

30. Komanoff (1982, pp. 215-217). Also, Komanoff (1976) found
that units Operated by AEP had superior capacity factor performance.
Capacity factor is a power plant’s actual generation as a percentage Of
its maximum possible generation, over a period of time.

31. Perl (1983, p. 2).

32. Ibid., p. 11.

33. Stewart, Op. cit.

145

34. French and Haddad (1980, p. 682) provide an excellent
discussion of the NERC (formerly EEI) availability data.

35. Joskow and Rose (1985, p. 20).
36. Ibid., pp. 23-24.

37. Schmalensee and Joskow (1986, p. 302) note this problem with
respect to unit heat rate, but they fail to recognize that the same
problem applies to the equivalent availability variable.

38. Ibid., pp. 299-300.

CHAPTER 3. DEVELOPMENT OF A GENERATING UNIT CONSTRUCTION COST AND
RELIABILITY MODEL

39. As noted in more detail later in this chapter, steam pressure
conditions have been the main avenue for improved thermal efficiency Of
coal-fired units built Since 1960.

40. Cowing and Smith (1978) provide an excellent review of the
econometric analyses of steam-electric generation. In their survey they
conclude '. . .the physical attributes Of the capital equipment appear
to constrain the ability of the economic agent in ensuing allocation
decisions, both those directly involving capital’s services and those
associated with labor and fuel.“

41. For an excellent discussion of how labor inputs are more
dependent on unit design characteristics than the level Of output, see
Bushe (1981, Ch. 4).

42. Komanoff (1976) also notes that a unit’s heat rate is
dependent to some extent on the level Of utilization of the unit since
heat loss is fairly constant at any load.

43. Stewart (1979), Cowing (1974), and Bushe (1981) provide good
discussions of the need to consider unit size and thermal efficiency as
characteristics of generating equipment. Houldsworth (1986) notes the
need to treat reliability as another attribute Of baseload units.

44. As noted on pages 4 and 5, the incremental heat rate Of a
generating unit is dependent on the level Of generation. It was noted
that there are a fairly large number of possible forms (see Figure 3-1)
for the incremental heat rate. He assume the incremental heat rate form
is represented by (d) or (e) and that the baseload units Operate under
normal load conditions. As a result, the amount Of fuel required is
generally proportional to the rate at which the unit generates
electricity.

45. Labor costs are ignored in our anlysis. This is unlikely to
bias our results for a couple of reasons. One, labor costs make up only
about 10% Of a unit’s generation costs. Two, labor input is dependent

146

on the engineering characteristics of the unit. But we are restricting
our analysis to baseload coal-fired units so any differences in labor
input requirements are likely to be minimal.

46. Everything else equal, supercritical units are 2 to 3 percent
more efficient than subcritical units with pressures of 2400 PSI [see
French and Haddad (1980) and Joskow and Schmalensee (1987)]. Both
studies also find that unit size has almost no impact on thermal
efficiency once steam pressure conditions are considered.

47. These relationships will be discussed in more detail in the
following section of this chapter.

48. See footnote 6 and pages 4 and 5 in this chapter for
additional detail.

49. See DOE (1985), pp. 38-39, Komanoff (1974, Ch. 7), and Vardi
and Aui-Itzhak (1981), pp. 20-21.

50. DOE (1985), pp. 30-31.
51. See Joskow and Rose (1985) and Bushe (1981, Ch. 3).

52. This was recognized as a possible source of bias by Joskow and
Rose (1985).

53. Ling (1964), Bushe (1981), Hills (1978), and Cowing (1974)
take no account Of variations in unit reliability across unit size or
steam pressure conditions.

54. See Komanoff (1974), Joskow and Schmalensee (1987), and Perl
(1982).

55. See Perl (1982) and Joskow and Schmalensee (1987).

56. See Joskow and Schmalensee (1987) and Schmalensee and Joskow
(1986).

57. Load-following is an intentional reduction in the level of
generation (power) due to low demand for electricity.

58. A logit transformation of the reliability variable was also
used for estimation. A logit transformation restricts the variable to
the 0-1 interval. The results are not reported since they are virtually
the same as those for the reliability transformation given above.

59. DOE (1985), pp. 38-39.

60. See Kmenta (1986), pp. 346-350.

61. See Kmenta (1986), pp. 357-361.

62. See Kelejian and Halker (1989), pp. 314-315.

147

63. See Appendix A for details of the deflation process. I did
not amortize the construction cost of a unit over its life since it is
assumed that all units have equal life expectancies. Thus, amortizing a
units’ construction cost would make no difference. The assumption of
equal life expectancies for different types and sizes Of baseload
coal-fired generating units is a standard industry practice. See EPRI
(1986).

64. Hhitman, Requardt, and Associates, Han dy-thtman Inga; of
Baltimore. Hhitman, Requardt and
Associates, 1986.

65. See Bushe (1981, Ch. 3), Smith (1977), and Joskow and Rose
(1985) pp. 3-4.

66. See Joskow and Rose (1985) and Bushe (1981, Ch. 3).
67. The data was made available to US by S.M. Stoller Corp.

68. It should be noted that no units were excluded due to poor
reliability. Units were excluded only if they were not observed during
their second through fourth year of commercial Operation.

69. Joskow and Rose (1985) specify the relationship between
average construction cost and size to be Cobb-Douglas. As a result, the
cost function has a constant elasticity Of unit cost with respect to
size. They state that more flexible specifications do not improve or
Change the results in a significant manner. Our results, using the
Cobb-Douglas specifications, are presented in Tables 8-1 through 8-3 in
Appendix 8. He prefer to estimate more flexible specifications that
allow the elasticity of average construction cost with respect to size
to vary with size and to change sign.

70. These parameter estimates are presented in Tables 8-4 and 8—5
in Appendix 8.

71. Alternative specifications of the reliability equation were
estimated, but made little difference in the general estimation results.
Also, introduction of other explanatory variables like a quadratic size
term cause some problems with multicollinearity.

72. Compare the results in Tables 3-1 and 3-6, Tables 3-2 and 3-7,
Tables 3-3 and 3-8, and Tables 3-4 and 3-9.

73. See Appendix C for details on how the unit cost elasticities
and their statistical significance were calculated.

74. The same specification Of the reliability equation was
estimated with all four specifications of the cost function.
Alternative specifications of the reliability equation were estimated,
but they did not change the results in any important way.

148
CHAPTER 4. AVERAGE CAPITAL COSTS AND THE UNIT SIZE-RELIABILITY TRADEOFF
75. For an excellent discussion, see Trebing (1981), pp. 369-385.
76. See Joskow and Schmalensee (19868) and Johnson (1985).
77. For a more detailed discussion, see Johnson (1985), pp. 47-48.
78. See FERC (1988).

79. Good surveys of state bidding programs are provided by Meade
(1987) and Nagelhout (1988).

80. See Nagelhout (1988) and Halker (1989).
81. See Rome (1988), p. 9.
82. See Halker (1989), p. 35.
APPENDIX 8
83. He used the same specification of the reliability function that

was used in Chapter 3. (See the parameter estimates presented in Table
3-10.)

BIBLIOGRAPHY

BIBLIOGRAPHY

Barzel, Y. ”The Production Function and Technical Change in the Steam-

Power Industry,” Journal at Balitigal Ecanomy, April 1964, 72, pp.
133-150.

Bernhardt, I. and Jung, 8. S. “The Interpretation Of Least Squares
Regression with Interaction or Polynomial Terms," Reyiew 9f
Economigs and Statistics, August 1979 61(3), pp. 481-483.

Burness, H.S., et al. ”Scale Economies and Reliability in the Electric

Power Industry,” lhg Energy Journal, January 1985, 6(1), pp. 157-
168.

Bushe, D. ”An Empirical Analysis of Production and Technology Using
Heterogeneous Capital: Thermal Electric Power Generation“
Unpublished Ph.D. Dissertation, New York University, 1981.

Corio, M.R. 'th Is the Performance of Electric Generating Units
Declining?,' Egblig Utilitias Egrtnightly, April 29, 1982, pp. 25-
30.

Cowing, T.G. ”Technical Change and Scale Economies in an Engineering
Production Function: The Case Of Steam Electric Power,” Jag:aal_af

Industrial.£conomios. December 1974, 23(2), pp. 135-152.

Cowing. T.G. and Smith, K.V. "The Estimation of a Production
Technology: A survey of Econometric Analysis Of Steam Electric

Generation,” Land_£tgaamits, May 1978, 54(2), pp. 156-186.

Electric Power Research Institute, l A m nt Gui ,Vol.1,
EPRI P- 4463- SR, December 1986.

om l i i r c stin
Be liability and Availability for Electric Power Generatingtynits,
EPRI AP- 2693, October 1982.

 

Ellis, R.P. and Zimmerman, M.8. ”Hhat Happened to Nuclear Power: A
Discrete Choice Model of Technology Adoption," lhe Revjaw of

Economics_and_§tatisticsY May 1983. PP- 234-242-

Farber, S. ”Buyer Market Structure and R&D Effort: A Simultaneous
Equations Model," h R vi w 0 O m c n Stati , August
1981, Vol. 68, No. 3. PP. 336-345.

149

150

Federal Energy Regulatory Commission, Bagglatign s figyarniag Bidding
Erggrams, Docket No. RM88- 5- 000, March 1988.

Fuss, M.A. ”Factor Substitution in Electricity Generation: A Test of
the Putty-Clay Hypothesis,“ in Ergduction Economies: A Dual
, eds. M.A. Fuss and D.L.

W905
McFadden, Amsterdam: North Holland Publishing CO., 1978.

Galatin, M. Eggagmigs gf Sgala and Iaghaglggigal Changa in lharmal

Egygr_§engratign. Amsterdam: North Holland, 1968.

Gill, G. S. "Omitted Cross- Sectional Effects in Measurement of Economies

of Scale in Electricity Generation,” in Etg_gmgtr1g_§tudjgs_1n
m n ,Ed. by G. S. Maddala, H. S. Chern, and
G.S. Gill, 1978.

Haddad, S.Z. and French, R.X. "The Economics Of Reliability and Scale

in Generating Unit Size Selection,” e i Of er n
RQ!§£_£9ntgrgaga, 1980, Vol. 42, pp. 680-686.

Haessel, H. “Measuring Goodness of Fit in Linear and Nonlinear Models,"

Sggtharn Eggngmig Jogrgal, January 1978, 44(3), pp. 648-652.

Houldsworth, M.A. “Abstract Scale Economies and Unit Availability in
Steam-Electric Generation: A Nonhomogeneous Capital Approach,"
Unpublished Ph.D. Dissertation, Michigan State University, 1986.

Hsiao, Cheng. Analysis at Banal Data, Cambridge University Press, 1986.
Huettner. 0. Plant Size. W

Raggirgmgnts. New York: Praeger Press, 1974.

Johnson, L.L.

WW
Qnngrtunitias_aad_£rgtlams, Rand Corporation, March 1985.

Johnston, J. Eggngmatrit_flathggs, McGraw-Hill, Inc., 1984.
Joskow, P.L. ”Productivity Growth and Technical Change in the
Generation Of Electricity," Ina Enargy Jgurnal, 8(1), pp. 17-38.

Joskow, P.L. and Rose, N.L. ”The Effects of Technological Change,
Experience, and Environmental Regulation on the Construction Cost
of Coal-Burning Generating Units," h n rnal Econ mic ,
Spring 1985, 16(1), pp. 1-27.

Joskow, P.L. and Rozanski, G.A. ”The Effects of Learning by Doing on

Nuclear Plant Operating Reliability,” I09 Reviaw of Econgmigs and
Statistits, May 1979, pp. 161-168.

Joskow, P.L. and Schmalensee, R. Markats for Eowar; An Analysis at
Wham. Cambridge: MIT Press. 1983.

151

and . 'The Performance of Coal-Burning
Electric Generating Units in the United States: 1960-1980,”
W

 

 

, 1987, Vol. 2. pp. 85-109.

and . “Estimated Parameters as
Independent Variables: An Application to the Costs of Electric

Generating Units," ddurna1_dt_figdndnatrits, 1986, Vol. 31, pp.
275-305.

 

 

an . ”Incentive Regulation for Electric

d
UtilitiesY" omm. 1986. Vol. 4. DP- 1-49-

Kelejian, H.H. "Two-Stage Least Squares and Econometric Systems Linear
in Parameters but Nonlinear in the Endogenous Variables," dgurnal
, June 1971, 66(334), pp.
373-374.

Kelejian, H.H. and Halker, E.0. ntro ct to nometri M thod ,
Harper & Row, 1989.

 

 

Kmenta, J. Elamants df Eggnomatrigs, Macmillan, New York, 1986.
Komanoff, C. ngar Elant Cost Esgalatinn, New York, N.Y.: Van Nostrand

Reinhold Co. Inc., 1981.

. £guar_£1ant_£arfdrnanga, Council on Economic
Priorities, 1976.

 

Komiya, R. ”Technological Progress and the Production Function in the

United States Steam Power Industry," Rayigu_dt_£gdngmi§s_and
Stat1§t1g§,May 1962, 44, pp. 156- 166.
Ling, S. on l tr we Gen ra n

Industry, Amsterdam: North- Holland, 1964.

Loose, V.H. and Flaim, T. "Economies of Scale and Reliability: The
Economics of Large Versus Small Generating Units," ne tem

Mm. 4042). 1980. PP- 37-55-

Maddala, G. S. Linitad Danan dan nt and Qualitatiyg Variables in
Etdndmgtrits, Cambridge University Press, 1983.

Martin, S. "Advertising, Concentration, and Profitability: The

Simultaneity Problem," Ing_Ba11_ddurnal_dt_£gdngnigs, Autumn 1979,
Vol. 10, NO. 3, pp. 639-647.

Mead, H.R. ”Competitive Bidding and the Regulatory Balancing Act,"
29W. September 17. 1987. 99- 22-30.

Mooz, H.E. ”Cost Analysis of Light Hater Reactor Power Plants," The
Rand Corporation, 1978.

152

Nagelhout, M. “Competitive Bidding in Electric Power Procurement: A

Survey Of State Action,“ Eub1ig_dtilitigs_fgrtnigntly, March 17,
1988. PP. 41-45.

Perl, L. “The Current Economics of Electric Generation from Coal in the
U.S. and Hestern Europe," National Economic Research Associates,
October 1982.

Romo, C. "Independent Power and the Alligators in the Moat," Eunlit
Vtilities.£9rtnigbtle November 24. 1988. 99- 8-9-

Smith, B.A. n v
1219, MSU Public Utility Papers, 1977.

Stewart, J.F. ”Plant Size, Plant Factor, and the Shape of the Average
Cost Function in Electric Power Generation A Nonhomogeneous

Capital Approach.” Bell_JourDal_9£_EsonomicsY 1979. 10(2). PP-
549-555.

i wer ener t n 95 —

Trebing, H.H. "Motivations and Barriers to Superior Performance under
Public Utility Regulation," in Erddugtiyity Maasuremant in
Regulated Industries, eds. T. Cowing and R. Stevenson, Academic
Press, 1981, pp. 369-394.

Vardi, J. and Avi-Itazhak, B. Elagtrig Engrgy Ganeration, The MIT
Press, 1981.

Halker, M.A. “New Jersey’s Competitive Bidding System-An Attempt at a
Balanced Energy Supply Policy," li i n Y
February 16, 1989, pp. 34-39.

Hills, H.R. "Estimation of a Vintage Capital Model for Electricity
Generating," Reviau gf Egonnmig Studias, Vol. 45, October 1978.

Zimmerman, M.B. "Learning Effects and the Commercialization of New
Energy Technologies: The Case Of Nuclear Power,” Ball_dgurnal_gf
Egdngmics, Autumn 1982, 13(2), pp. 297-310.

U.S. Department of Energy. te nts it l O for o -F' ed
Edygr_£lants, DOE/EIA-0479, October 1985.

    

STRTE U V

Intrillllltiirllllit

£3

 

MICHIGAN
NW
531 I?