Effects on relative efficiency in electric power generation due to environmental controls

Rc~ounn and Energy 8 (1986) 167484. North-Holland

EFFECTS ON RELATIVE EFFICIENCY IN ELECTRIC POWER GENERATION DUE TO ENVIRONMENTAL CONTROLS R. FARE, S. GROSSKOPF and C. PASURKA’ SouthemIllinois Uniomiry, Carbondale, IL 62901. USA Rcaived June 1985. final vcrsioo received November 1985

IO this paper. we model the effects of environmental controls when these arc viewed as rcxtrictiog diiposal of outputs, We specify an unregulated and a regulated multi-output technology satisfying strong (free) disposability of outputs and weak (restricted) disposability of outputs, respectively. By comparing the relative productive efficiency of firms using restricted and unrestricted technologies we derive a measure or index of the Act of environmental regulation. We employ recently developed prolpammiog models to calculate this measure of the cost of environmental controls for a sample of steam electric plants in the U.S.

1. Introduction The Clean Air Act Amendments of 1970 initiated the process of setting national ambient air quality standards, which were to be met by 1975 [Lave and Omcnn (1981)]. Water quality standards were established by the Water Quality Act of 1965. Since these and various state regulations are directed mainly at controlling industrial sources of air and water pollution, the impacts of environmental regulations on the performance of the economy have been a source of concern. These regulations set technological standards intended to limit the level of various emissions including sulfur oxides, thermal pollution, nitrous oxides and particulates. As a result one of the industries most severely impacted by environmental legislation is the electric utility industry, particularly those plants relying on coal to generate electricity. The purpose of this paper is to determine the impacts of environmental regulations on the relative efficiency of steam electric utilities. In particular, we focus on the indirect costs of environmental regulations rather than on the direct outlays for pollution abatement which are typically employed in determining costs of environmental controls. These indirect costs arise when regulations restrict the production process, resulting in a loss of effkiency or

*Theauthors arc gratefulfor commcnl~ 0165-0572/86/S3.50

@ 1986. Elrvicr

by Prof. V. Kerry Smith and a referee of this journal.

Science Publishers B.V. (North-Holland)

168

R. FZre

et al.. Effuimcy

in elecmhalpower genmtion

productivity relative to the unrestricted production possibilities.1 Previous studies by Kopp (1980) and Kopp and Smith (1981) suggest that enviroomental regulations significantly affect productivity/efficiency. Their conclusions have been corroborated by several other studies. Gallop and Roberts (1983) specified a translog cost function of the electric utility industry. Among its independent variables was a measure of regulatory intensity regarding sulfur dioxide emission standards. They found that from 1973 to 1979 sulfur dioxide emission restrictions reduced the average rate of productivity growth in the electric utility industry by 0.59 percentage points per year for constrained utilities. The studies discussed up to this point used a single output - electric power - in modelling and estimating the production process. Tran and Smith (1983, p. 40) generalized this approach by specifying a ‘translog joint output cost function’ which included air and water pollution emissions as additional outputs. They concluded that air and water emissions should be included among the outputs of electric utilities. In addition, they found that environmental regulations influence the optimal usage of inputs. In an earlier study, Thompson, Calloway and Navalinic (1977) specified a linear programming model of electric power generation, which assumed constant returns to scale, to calculate the impact of environmental regulations on the cost of producing electricity. The model had three types of inputs: fuel, water and capital. Its outputs consisted of electricity production and six different types of waste discharges to the environment: two air, three water, one land. Simulations were run to determine the effects of different emission standards and effluent taxes. They found that production costs increase from SL25/kWh to a maximum of SZ.OO/kWh because of environmental regulations.’ In this paper, we use a linear programming approach, but we do not impose constant returns to scale. In accordance with the results of Tran and Smith (1983) and Thompson, Calloway and Navalinic (1977), we specify a production technology which can be used to model the joint production of a good (electric power) and ‘hads’, i.e., emissions which are subject to environmental regulations. What is unique about our approach is that this technology can also distinguish between a controlled prc%ess (where bads are regulated, i.e., not freely disposable), or what we call ‘weakly’ disposable and an unregulated process in which all outputs, good or bad, can be disposed of costlessly by the firm (called strong disposability in this paper). The purpose of environmental control is to limit disposability of (bad)’ outputs. One can therefore, by modelling a production technology under both weak and strong disposability of outputs, determine the regulated and unregulated technologies, and consequently, determine the impact of environmental controls ‘Those restrictions will, of course. produce bcnclits in the form of rcduccd environmental damages. Our focus hcrc, however. is on the cost cffccts. ‘See ch. 6. The maximumis based on maximum restrictions on air and water quality.

R F&e et d.. E/Trriencyin electricd power gentration

169

on a technology. That is, unlike earlier stud& we can *directly’ model and measure the ‘indirect’ impact of regulations on the technology of our sample of steam electric plants. Specifically, we apply a multi-output Farrell (1957) measure to a sample of electric power generating firms under the alternate assumptions of weak and strong disposability of outputs. Using linear programming techniques with the observed data forming the best practice or frontier technology,3 we measure output efficiency for each observation relative to the frontier of the technology. By comparing these efficiencies for both the weak and strong disposability (froatier) technologies, the indirect impact of environmental control on relative output efficiency is calculated. Using these results we can calculate the opportunity cost of environmental control in terms of output foregone. In the next section we present the basic theoretical model which specifies the technology and introduces the efficiency measures to be used to calculate the impact of regulation. Section 3 contains a discussion of the data used in this study. We present the results in section 4 and a summary in section 5. 2. Specification of the technology and the efliciency measures

In the theory of production one distinguishes between weak and strong disposability of outputs. Typically, it is assumed that outputs are strongly disposable, i.e., that their disposal is ‘free’. In a regulated environment, that assumption is clearly not valid. Alternatively one can relax the strong disposability assumption to weak disposability. With weak disposability, an output is no longer necessarily freely disposable. Rather, disposal of an undcsirablc output, for example, imposes a cost in the sense that it is achieved by reducing the other outputs proportionately. More formally, weak disposability of outputs means that if an output vector u is obtainable with an input vector x, then all output vectors u, which are proportionally less than II, i.e.. u= 0 *u, 0s 04 1, are also obtainable with x. Strong disposability means that if u is obtainable with x, then so are output vectors u that are smaller than u, i.e., one can freely dispose of outputs. Before introducing the efficiency measures to be used in this study, we need to specify (piecewise linear) reference technologies which satisfy strong or weak disposability of outputs. Since we have multiple rather than a single output, the appropriate production form is the output correspondence rather than a single-valued production function. Following Fire and Grosskopf (1983) [see also Fire, Grosskopf and Love11 (1985), and Shephard (1970,1974)]. and given observations of inputs ‘The data points arc ‘enveloped’. The boundary fomcd by enveloping the da;a points is the best practice or frontier of the technology.

R.Fdre et al.. El/rrim in elmricai power gcnemtiom

170

xy and outputs I+ tbe strougly disposable output correspondeuce can be formed as follows. Let xrf denote the jth input used by firm i and let utl denote the jth output produced by firm i. Denote by M the matrix of observed outputs and denote by N the matrix of observed inputs. The strongly disposable output correspondence can be written as

where ZE @+ denotes the intensity vector familiar from activity analysis. This correspondence thus tells us how much output could be produced using x as the input vector. In order to determine output loss due to constrained or controlled output disposability, we need to construct a weakly disposable output correspondence. Following Shephard (1974) this can be done as follows: P”(x) =

1

uERm+:U=&?M,ZN~x, f

i=l

t*= 1,6E[O, l] .

(2.2)

I

To elucidate the diflerence between the output correspondences P‘(x) and P’“(x), consider fig. 1. In this figure, the output correspondence P’“(x), i.e., the output correspondence for which disposal of outputs is not necessarily costless, is bounded by the line segments OA, AB, BC and CO. The correspondence P(x) is bounded by EO, EB, BC, CD and DO. It is clear that P”(x)E P’(x), and if P”(x) is, as in fig. 1, a proper subset of p(x), then output may be lost due to lack of disposability. To see this, consider point B, and suppose that u, is a ‘good’ and u2 is a ‘bad’. If disposal of the bad is costless, that implies that the dotted line

Fig.

this

I

vector is a variable and serves to connect or

envelopthe data points in

M aad N.

R

Faireet d. E~uiency in eluwical power

gentration

171

segment (along which the bad is reduced from its original level at B) would be a feasible part of the technology. Ic on the other hand, disposal of u1 is not costless (say, for example, capital and labor must be used to dispose of it). then the segment EB is no longer feasible since the resources used to clean up u2 imply that production of OE units of ui are no longer possible given the reduced available inputs. Reduction of u2 is achieved at some cost in terms of ul, as along the segments BA or OA. To actually calculate the output loss due to environmental control, i.e., restricted disposability, we solve two programming problems, namely: W(x, u) = max 1

subject to (2.3)

F(x, u) =max 0 Bu=SrM,

subject to

rN$c,

,$i,,=l,

SE[O,l],

ZER:.

The first of the two problems maximizes potential output relative to the technology P(x), i.e., (2.1), while the second programming problem maximizes potential output relative to the technology P’“(x), i.e., (2.2). We note that in both (2.3) and (2.4) we do not force any restrictions on the technology in terms of returns to scale. Following Afriat (1972) we employ the restriction c z, = I which allows the technology to display increasing, decreasing or constant returns. That in turn implies that our efficiency measures are independent of the scale economies of the technology, i.e., the existence of increasing returns will not, for example, result in a plant being more or less efficient than a plant exhibiting constant or decreasing returns. To illustrate the programming problems (2.3) and (2.4), consider fig. 2. As in lig. I, the strongly disposable output technology in fig. 2, f’(x), is bounded by the line segments OE, EB, BC, CD and DO. The weakly disposable output technology, P’“(x), is bounded by the line segments OA, AB, BC and CO. Given an input-output vector say (x, u) with uoP”(x), the program (2.4) measures the maximum potential output proportionally obtainable given P’“(x). This maximum output is given by u-F(x, u), as shown in fig. 2. In a similar fashion, the problem (2.3) gives us the maximum potential output proportionally obtainable from the technology P(x). This maximum output is represented in fig. 2 by u - W(x, u). Thus an observation on the outer boundary of the relevant technology [like points B and C for P(x) and point A for P’“(x)] is dubbed ‘eflicient’, i.e., W(x, u) or F(x, u) are equal to unity for such observations. In words,

172

I

D

Fig. 2

efficient firms are producing the maximum achievable outputs given their inputs and the relevant technology. Again, the frontier of the technology is determined by the firms in the sample, not by some theoretical engineering standard or presupposed functional form. Returning to our problem of measuring the effect of environmental control on efficiency, the regulated technology would be the weakly disposable technology P”(x). The unregulated technology is represented by P’(x). Thus F(x, a) captures deviations from dfrciency in the regulated environment and W(x, a) captures deviations in efficiency in the unregulated environment.s Note that W(x, u) 2 F(x, u), i.e., it ‘contains’ any efficiency loss relative to the regulated technology. We would like to measure the loss of efhciency (output) due explicitly to environmental control regulations. That is we would like to capture the distance between point u-F(x, u) and u* W(x, u) in fig. 2. We call this measure C(x, u) which is calculated as

w,

4-

wx,u)lF(JG4.

(2.5)

C(x, u) measures the radial difference between u- F(x, u) and us W(x, u), i.e., output loss due to lack of disposability. After discussing our data set, we present calculations of W(x, u), F(x, u), and C(x, u) for each observation in our sample. 3. Data The data used in this study consist of 100 randomly selected steam electric utility plants in 1975. For each plant, information was collected on five inputs. The capital variable is installed generating capacity (in megawatts) ‘Thesemcasura woubd alsu capture any ‘X-incf@iency’ or managerial slack in a firm. They do not, however. capture inellicicncy due to allocative or price-related ‘mistakes’.

and the labor variable is the average number of employees [U.S. Department of Energy (1978)]_ The three remaining inputs are the amounts of different fuels used by the plant @J.S. Federal Energy Regulatory Commission (1979)]: (1) coal (1,000 tons). (2) oil (1,000 bbls.). and (3) gas (million cubic feet). The framework we use allows for substitutability of fuels. Not all of the plants in our sample, however, use all three types of fuels. Of the total, however, 78% of our observations are multifuel plants. Information was also collected on five outputs for each plant. The only ‘good’ output is net generation in million kilowatt-hours [U.S. Department of Energy (1978)]. The remaining outputs consist of four measures of pollution [U.S. Federal Energy Regulatory Commission (1979)]. Three types of air pollution emission (1,000 tons) are included: particulate matter, sulfur dioxide, and nitrogen oxides. The fourth type of pollution is a measure of the heat discharge in the water used by the plant. It is calculated by taking the product of (1) the average rate of withdrawal (cubic feet per second) and (2) the difference between the maximum temperature of the water at diversion and outfall during the peak winter month. Mean values of these variables are presented in table 1. Also included in table 1 are mean values of several other characteristics which are of interest for the steam electric utility case including vintage (initial year of operation, or year of last major change in equipment) and plant factor (plant capacity utilization factor for 1975). These characteristics will bc used in analyzing our results. We do not include them as inputs; the plant factor is, however, captured in the capital and net generation variables. 4. Results Using the linear programs and calculations described in section 2, three measures of efficiency were calculated for the 100 plants in our sample: W(x, u), F(x,u) and C(x,u). These are summarized in table 2. Relative to the unregulated technology, almost l/5 of the plants in our sample are ineficicnt, i.e. W(x, u)> 1. If, however, one uses the restriction or weakly disposable technology as the reference for efficient operation, only 3 plants are ‘X-inefficient’ [i.e., only three plants have F(x, u)> 11. Thus, if pollution regulations are ignored [i.e., W(x, u) is used as a measure of efticiency instead of F(x, u)], the loss of efficiency (‘X-inefficiency’ or organizational slack) of the plants in our sample would be overstated. It appears that, given the restrictions on disposability implied by the regulations, these steam electric plants are relatively effcient. Those regulations do, however, impose costs on the regulated plant. We can capture the opportunity cost of those regulations in terms of lost output due to lack of disposability of outputs imposed by the regulations. This is captured by our measure C(x,u). As seen in table 2, 15 of the 100 plants did

174

R. F&e

et 1. E~iiency ir, &ctricd

power generation

Variable

outpats Net generation’

Millions of hilowatt-hours

2333.89

218025

Particulate matter’

loo0 tons

281

7.50

sulfur diisideb

1000 tons

29.91

4790

Nitrogeo osidcb

1000 tons

Heat

(See text)

10.52

1158

809217

8225.30

lnstllkrlgcncrating in megawatts

60267

496.07

Average number of employees in 1975

116.07

78.93

fnputs

Capital’

capacity

Labof CoaP

WOOtons

574.53

925.27

Oilb

1000 bbls.

1183.13

1804.51

Gash

Million cubic feet

4047.86

0749.99

Plant vintage (initial year of operation)

YtSMS

1952

Plant factor (net generation capacity)

Percent of capacity

Churactcristics

44.83

12.90 16.70

‘U.S. Department of Energy (1978). bU.S. Federal Energy Regulatory Commission (1979).

lose output due to lack of output disposability. This loss ranged from a tenth of one percent to almost 48”/, of output. in order to get a bettor idea of the magnitude of this loss, we can calculate the potential output (net generation) lost due to inefficiency. Potential output can be calculated by multiplying actual output (net generation) by the calculated ctliciency measures in tabk 2. The loss of potential output due to incffkkncy would then be the diK&cnce between potential and actual output.6 These figures arc summarized in table 3. Based on the regulated reference technology existing in 1975, there was a loss of approximately 434 million kilowatt-hours due to ‘X-ineffickncy’ for our sample of ekctric utilities (this is referred to as FLOSS in table 3). The ~Ahernativtly. this can be eakulatal directly M (C(x. u)- I)+output

R. Fire et d. EJi&ency

in elecuical

power generation

Table2 ElkicncymeasuresSteam &ctric plants. 1975. Observation

: 3 4 5 6 7 8 9 10 I1 12 13 14 15 16 17 18

21 22 23 24 25 26 27 28 29 30 :: 33 34 35 36 37 38 39 40

w. l.OW 1.000 1.000 1.000 1.000 1.000 ::E 1.000 1.000 1.000 l.ooO 1.044 l.ooO l.WO 1.000 1.000 1.000 ::iZt l.000 1.000 1.006 Loo0 1.004 l.Oal woO 1.000 1.000 1.000 1.000 1.020 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.369

41 42 43 44 4s 46 47 48 49 50

1.002 l.ocQ 1.m mlo 1.040

::

l.ooO

moo

1mo

u)

W. ~1

c(x. u)

l.CNIO 1.000 l.WO 1.000 l.WO l.OC0 1.Mk-l l.CKlO 1.000 l.tXlO 1.000 l.OWl l.ooO 1.000 l.ooO 1.000 1.000 l.OW 1.000 l.OW 1.000 l.ooO l.m l.ooO l.m 1.000 l.aKl 1.000 1.000 1.000 l.m 1.01s l.ocQ 1.000 1.000 Loo0 Loo0 l.OMl 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 l.ooO l.OiM 1.000 l.ooO l.ooO

1.000 1.000 1.000 1.000 1.000 l.ooO 1.000 1.000 1.000 1.000 l.OCMl 1.000 1.044 l.@JO l.ooO l.O@J 1.000 1.000 t:iE 1.000 1.000 l.CGO l.ooO 1.004 1.000 1.000 1.000 1.000 1.000 1.000 1.005 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 ml0 1.369 1.m 1.002 1.000 LOW 1.000 1.040 1.000 1.000

175

R FBr et d.. Eff~ieuy

176

53 54 55 56 57 58 : 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 lb

77 78 79 80 at 82 83 84 85 86 a7 88 : 91 92 93 94 95 96

97 98 99 100 Arithmetic mean (Standard deviation) bfcdian Source: See section 3.

in ektricd

1.000 1.001 l.ooO 1.252 1.000 l.ooo l.ooo l.ooo 1.002 1.124 l.ooO 1.000 1.003 l.ooO l.ooO 1.@30 1.000 l.ooo l.ooo l.ooo l.ooo 1.001 1.000 l.ooo lam I.ooo l.ooo l.cm l.ooO 1.000 1.000 1.000 1.346 Iam l.ooo l.WO l.ooO 1.000 1.000 1.010 l.ooo l.ooo 1.000 1.000 I.000 l.ocm 1.481 1.000 I.017 (0.074) I.000

power generation

l.ooo l.ooO l.WO lAXI l.oaI l.ooO l.ooO 1.000 1.000 1.025 l.mo l.ooa l.ooo l.cao 1.m l.aoo l.ooo LO@3 1.000 l.ooo l.ooo l.ooo l.oal l.ooo l.ooo Kii 1.000 Loal l.ooo l.OW l.ooo 1.330 I.000 I.000 1.000 1.000 l.ooo 1.000 1.000 l.oocl l.ooo l.ooO I.000 l.ooo l.ooO l.ooo l.ooO

l.ooo 1.001 l.ooo 1252 l.OCO l.CHlO l.OW 1.000 1.002 1.096 1mo 1.m 1.003 1.000 1.000 1.000 l.ooO l.ooo l.ooil l.ooo l.ooo I.001 Moo l.ooO l.cm 1.000 t:kZ l.ooo l.ooa 1.000 l.ooo I.012 l.ooo l.ooO l.ocm l.oofl 1.000 1.000 1.010 mm 1.000 1.000 1.000 1.ooo 1.m 1.481 1.000 1.013

(A:; l.ooo

(0.W l.ooO

R hire

et aL. E/limcy k l&cwical power generation I

I

Table3 Lossofnetgatcmiugpowerductoind5cicacy (in millioo kilowatt-hours). Obcrvation

wwss

FWSS

O.OW

o.am

: 3 4 5

EE :El 0.000

i 8 9 10 11 12 13 14

0.000 O.WO O.ooO

0.m 0.000 O.ooO 0.000 O.ooO 0.000 O.ooO

15

16 11

i-iii oiao 93.830 0.000 0.000 O.ooO O.WO

18 : 21 22 23

:Ei OiOO ZE 0.366 0.000 2.458 0.000 i-E OiloO

44

45 46 47 48 49 50

0.000

:z

oioo 0.000

z-iii oioo O.WO O.OW

El :iE O.ooO

O.OCUl

O.WO 0.000 O.ooO

Ei oioo :E O.ooO O.OMl

i-E

42 43

CWSS

EE oioo

O.ooO O.CUlO

:tE

93.830 :z OiIOo 0.000 :Ei Ez OilOO 0.366 O.WO 2.458 O.ooO ~~ OioO O.WO 0.000 4.945

13:873 0.000 0.000 O.CNlO :z O.OW 0.000 0.000 :iE O.CNKl 0.000 :ii 0:ooo O.tXJO o:OoO O.ooO O.ooO :E O.ooO 0.000 0.000 O.ooO 0.000 633.278 O.CKUI 63i.g oiOO O.CUKl 0.000 2.420 0.000 2.420 0.000 0.000 :E E-i 0.000 0.000 OiOO 56.444 0.000 56.444

i:Z iEz ::Zi

rn

_ 178

observation

wwss

FWSS

51

aam o.ow

:: 54 55

O.OWl 0.000 O.CKXl 0.000

$ 58

276284 0.000

59 0.000

: :: 64

65 66 67 68 69 70 71 72 73

Ki oioo 0.000 0.000

2572 412799 84.694 O.ooO O.WO 0.000 0.000 4.257 o.ooo

~~ O.OW O.fXiO 0.000 OSXJO 0.000

3.655 0.000 0.000 0.000 0.000 O.OW O.ooO O.ooO 0.000 0.000

Ez iEz 0.000 0.000 O.NlO 0.000 O.ooO 0.000

Etz 0:ooo 0.000 0.000

o.ioo

351.091 335.049 0.000 0.000 O.ooO 0.000 90

91

98 99 100 Total

cwss 0.000 O.OW

1.402 O.ooO 276.284 O.OW :Ei OiOO 2572 319.978 0.000 0.000 4.257 OSKKI O.OW O.OWl O.OCQ O.OW :E o:oOo 3.655 0.000 0.000 0.000 0.000 :Ei 0:ooo 0.000 0.000 O.WO 12.061

ii% o:OoO 0.000 O.ooO Kz 0.000 O.MlO OiaO 0.000 20.738 20.738 O.ooO 0.000 0.000 0.000 0.000 O.OCKl 0.000 0.000 %i 0.000 oiloo %z 0.000 OMlo O.OW OiOO 187.713 187.713 0.000 0.000 0.000

2068.20 433.620 1622.400

R F&c et d.. E/l’lcLncy

in elecrrical pwa generation

179

5,

output due to the regulation tie_ due to C(x, a)] is approximately 1,622 million kilowatt hours in total, or nearly four times the loss due to F(x,u) or ‘X-ineIliciency’. We would expect that loss to become larger after 1975 as the federal environmental regulations were implemented and amended. Some additional empirical evidence as to the sources of observed inelllcieocy in our sample is summarized in table 4. Included are Pearson correlation coefficients calculated for the efficiency measures, emissions and input usage of the plants in our sample. Of particular interest is the source or cause of output lost due to lack of disposability. The cause or lack of disposability need not be one particular emission or output, but rather could be due to any combination. Based on the correlation coefficients, we find that all of the outputs are negatively related to C(x,u) although none are significantly related to C(x, u) at conventional levels of significance. Note that thermal water pollution (heat) is negatively and significantly related to W(x,u) at the 90% level. This suggests that the greater is heat abatement (i.e.,

Iogg

in

Table Pearson Variable

correlations. wx.

Net generation

4

u)

9%

u)

C(x. u)

-0.11888 (0.2388)

- 0.06039 (0.5506)

-0.10326 (0.3066)

- 0.00302 (0.9762)

0.03151 (0.7557)

-0.01938 (0.8482)

Sulfur dioxide

-0.08181 (0.4184)

-0.04883 (0.6295)

-0.06715 (0.5068)

Nitrogen

-0.09435 (0.3504)

- 0.04960 (0.6241)

-0.08102 (0.4229)

-0.16736 (0.0960)

-0.081 I7 (0.4221)

-0.14723 (0.1438)

Particulnle

matter

oxides

Heat

0.02208 (0.8274)

(0.9623)

0.02676 (0.7915)

(0.7514)

-0.04750 (0.6388)

0.06001 (0.5531)

coal

-0.05019 (0.6199)

-0.02823 (0.7804)

-0.04212 (0.6773)

Oil

-0.03199 (0.7520)

-0.03968 (0.6951)

-0.01694 (0.8733)

Gas

-0.03457 (0.7328)

0.01394 (0.8904)

Vintage

-0.13881 (0.1684)

-0.19988 (0.0462)

- 0.05465 (0.5892)

Plant factor

-0.17938 (0.0741)

-0.07661 (0.4487)

-0.16275 (0.1057)

Labor Capital

-0.04603 (0.6493) Q

180

R F&e et al.. i$%ency

in ekcttical

power iy~doa

the iower is heat), the greater is the loss of output relative to the unregulated technology. The fact that thermal pollution bears the only significant relationship to efficiency losses may have to do with the fact that water pollution regulations have been in place longer than air emission regulations and were perhaps more effectively enforced by 1975. Approximately 75% of our sample was subject to some type of thermal pollution regulations at the state level in 1975 [Patterson (1977)-j. The present federal standards did not go into effect until after 1975. As further evidence, we calculated means of the dficiency measures for plants located in states in which there were numerical thermal pollution regulations in 1975 versus those plants in states in which such regulations did not exist. These figures appear in table 5. According to these results, plants in states with such regulations were more eficient than plants in states without regulations. This apparent anomaly may be due to adjustments in the technology to avoid thermal pollution. We note that net generation is negatively related to all of the efficiency measures (at relatively low levels of significance). This suggests that plant size may have an effect on efficiency, i.e., bigger plants lose less output (in percentage terms) due to regulations than small plants. In an earlier study, Pittman (1981) found a positive correlation between pollution control intensity and plant size for a sample of paper mills. He argues that this larger size would result in a less competitive environment and a loss of eficiency. Based on our cross sectional evidence, this seems not to be the case here. In an’ earlier study, Kopp (1980) found that the impact of environmental controls on electric utilities varied strongly by vintage and by type of plant, i.e., whether base or peak load. To see if our results corroborate those findings we disaggregated the sample by plant factor and vintage. Following Dhrymcs and Kurz (1964) and Kopp (1980) WC grouped the plants in our sample according to the vintage classes: 1918-1937, 1938-1945, 19461950, 1951-1954, 1955-1959, 1960-1966, 1967-1974.’ Since we had no direct information as to whether individual plants are base, cycling or peak load, we grouped plants according to plant factor (percent of capacity generated in 1975). Two different definitions of base load were used: (1) plant factor of 70% or more and (2) plant factor of 50% or more. Mean values of the eficiency measures for these groups appear in table 5. Looking lirst at vintage categories, we find that the oldest vintage class is on average the least efficient in terms of ‘X-inefflciency’. The vintage class which was most strongly affected by lack of disposability was the 1955-1959 vintage. In his study, Kopp (1980) found the largest efficiency impacts of his proxies of regulation to be on 1960-1966 vintage plants. These results are not, however, directly comparable - Kopp used only one output in calcu‘The ‘vinlage’

year refers to the first year of operation

of the plant.

R. FZrelI al.. Ef_ciencyin

ektrical

power

181

genemtioa

Table 5 Group mcaaa (staadard dcviatious).

Group

wx

Base load

1.0003 (0.009) 1.019 (0.078)

(plantfactor 2 7O%J Other (plant factor < W$ Base load (plant factor 2 50%) Other plant factor < WA)

a)

0x, u)

C(x* a) 1.0003 (0.009) 1.015 (0.069)

1.001 (0.006) 1.027 (0.093)

(Kg

1.049 (0.113) 1.007 (0.015)

Nut&r of observations 10 90 38

1.006 (0.042)

1.001 (O.Ow 1.021 (0.083)

1.025 (0.092)

1.023 (0.070)

13

1.m (O.Otm

1.007 (0.015)

8

Loo02 (0.007)

Loo02 (0.007)

19

(Ezl)

1.021 (0.087)

1.021 (0.087)

18

(EZ, (

A:E, (E,

12

(E, 1.007 (0.028)

I.002 (0.006)

1.005 (0.021)

20

,A:&

(EZ)

62

Vintage 1918-1937 1938-1945 1946-1950 1951-1954 1955-1959 1960-1966 1967-1974

10

Ownership

Public (0)

1.046 (0.144)

1.001 (0.005)

Private (I)

1.013 (0.060)

(kg)

I1 89

Stute therm1 reguluriuns Yes (0)

No(I)

1.017 (0.073) 1.018 (0.080)

1.005 (0.037)

1.012 (0.062)

79

1.018 (0.080)

21

lating efficiency and also calculated efficiency for an earlier time period, 19694972. The only vintage group that was not adversely affected by lack of disposability in our sample was the group of most recent vintage plants. Turning to the means of the effciency measures by type of plant (table S), under both definitions of base load, base load plants were not as adversely

182

R. Fire et al.. Eflikacy in dearid

powergenmnion

affected by lack of disposability as were other plants. Since the relatively new vintage plants are more likely to be base load, this result is not too surprising By way of sutnmary statistics, correlation coe&ients for vintage and type of plant appear in table 4. Newness and relatively intensive use of capacity are associated with unproved efficiency, although that correlation is significant only for vintage with F(x, u) and plant factor with W(x, u) at the 90% level of confidence. We note that some of these electric utility plants are subject to regulations other than those dealing with emissions. Of particular impact is this industry are rate of return regulations. Insofar as these regulations affect relative input usage (for example: through an Averch-Johnson-type effect), privately owned utilities may differ from publicly owned utilities with respect to ‘X-inefficiency’. Mean values of the efficiency measures for publicly and privately owned plants appear in table 5 and suggest that the publicly owned plants in our sample are, on average, slightly less ‘X-inefficient’ than the privately owned plants. On the other hand, as a group the publicly owned utilities were more adversely affected by lack of disposability on average than the privately owned plants. It would be interesting to know if the differences in efficiency observed by vintage, plant factor, state thermal regulations, and ownership are significant. To shed some light on these issues we have employed several methods to test the hypothesis that characteristics such as ownership affect efficiency. The tests include a simple analysis of variance test and two non-parametric tests: the two-sample median test and the Kruskal-Wallis test [for details, see Siegel (1956)]. The analysis of variance method is based on the assumption that our measures are normally distributed, which may well be overly restrictive, given the nature and method of calculation of the efftciency measures. We employ the non-parametric tests precisely because the efficiency measures as calculated here are not likely to satisfy the standard normality assumptions inherent in standard statistical tests. The two-sample median test is a non-parametric test which can be used to determine whether the central tendency of the measures (in particular the median) is different across samples. The Kruskal-Wallis test is another non-parametric test which is based on the entire distributions of the samples to be compared and is, therefore, more powerful than the median test. The test statistics for these tests appear in table 6. Since our focus is on the effect of environmental regulations on e&iency we present the test statistics only for the C(x, u) measure. The null hypothesis is that the plant groups are equally efficient. Based on a 90”/, level of confidence, the only signifiamt differences appear for plants with generation equal to or greater than so”/, of capacity versus those with plant factor less than soo/ and for public versus private ownership. For plant factor, only the non-parametric tests show a

R.Fie et al.. E~iiency in electrictdpower gemmtion

183

Table 6 Test statistics, for C(x, u) by type of plant.*

BiUC-lOad

vs. other@ Base-load vs. othef

KN.%kal-Wallis

:Rob>F)

$rob>fi

$rob>fi

0.514 (0.43)

0.295 (0.59)

2100 (0.15)

2.966 (0.09)

3.23 (0.07)

(E)

6.12 (0.41)

6.31 (0.39)

0.13 (0.72)

(&

0.07 (0.79)

By vintaged Regulation’ Ownership’

of variana

Median test

Analysis

2.89 (0.097)

0.11 (0.75)

7he null hypothesis is that the value of C(x,u) is the same for the two groups of plants. bBase-load delined as plant factor 2 70”/. in 1975. ‘Base-load delined as plant factor 2 50% in 1975. dThe vintage groups used were the same as those used by Dhrymes and Kun (1964) and Kopp (1980). Specilkally, the vintages (first year of operation) were grouped as follows: 1918-1937. 1938-1945. 194Gl950. 1951-1954, 1951-1959.

1960-1966. 1967-1974. ‘Plants in states with numerical thermal pollution regulations versus plants in states without such regulations. ‘Publicly owned versus privately owned plants.

significant difference. For the ownership case, the plant groups exhibit significant difference only in terms of variance. From these results we conclude that non-bascload plants and publicly owned plants are most affected in terms of eff%icncy by the restrictions on disposability inherent in environmental regulations. 5. Summary

In recent years there has been renewed concern that government rcgulations reduce the efficiency of the industrial sector. ‘Coal-fired’ steam electric plants have been particularly hard hit by a variety of regulations. In this paper we focussed on the impact of regulations on electric utilities. We modeled the regulatory environment as a restriction on the technology. Spccilically, regulations were assumed to reduce the disposability of outputs (particularly ‘bads’ such as sulfur dioxide). We then captured the opportunity or indirect cost of those regulations as the difference in output when disposability is free (i.e., unregulated) and when disposability is restricted. We calculated this opportunity cost for a sample of 100 electric utilities in

184

R. F&e et al.. Effuimcy in electrical pwer generation

1975. Even though most of the current federal pollution control regulations had not become effective at that time, we found that lack of disposability ‘cost’ an average of roughly 16 million kilowatt-hours in lost potential output for each plant in our sample. This is in addition to any direct outlays on pollution control equipment, for example, suggesting that simple outlay measures of the costs of pollution control understate the social cost of improving environmental quality. We also found evidence that the largest impact was oa publicly owned plants and on plants with plant factors less than so”/, which are probably off-peak or cycling plants. References Afriat. Sidney. 1972, Efficiency estimation of production functions, International Economic Review 13.568-598. Dhrymes, Phoebus J. and Mordecai Kun 1964, Technology and scale in electricity generation, Econometrica 32, no. 3.299-315. Fire. ROB and Shawna Grosskopf, 1983, Measuring output eliiciency, European Journal of Operational Research 13.173-179. Fiire. ROB, Shawna Grosskopl and C.A. Knox Lovell. 1985, The measurement of elliciency of production (Kluwer Nijhoff, Boston, MA). Farrell. Michael, 1957, The measurement of productive elliiency. Journal of the Royal Statistical Society, Series A. CXX, 253-281. Gallop. Frank M. and Mark J. Roberts, 1983, Environmental regulations and productivity growth: The case of fossil-fueled electric power generation, Journal of Political Economy 91. no. 4. 654-674. Kopp. Raymond, 1980. Energy residuals and inefliciency: An engineering-econometric analysis of environmental regulations. in: John Moroney, ed.. Advances in economics of energy and resources. Vol. 3. 199-219. Kopp. Raymond J. and V. Kerry Smith, 1981. Productivity measurement und environmental regulation: An engineering-econometric analysis, in: Thomas Cl. Cowing and Rodney E. Stevenson, eds.. Productivity mcasuremenl in regulated industries (Academic Press, New York) 249-28 1. Lave. Lester B. and Gilbert S. Omenn. 1981. Clearing the air: Reforming the clean air act (The Brookings Institution, Washington, DC). Potterson. James W.. 1977. Directory of federul and state water pollution stundards. Illinois Institute lor Environmental Qtmlity. Chicago. IL: also in: Jack Golden. Robert P. Ovellette. Sharon Saari and Paul N. Cheremisinolf, edr., Environmental impact data book, 1979 (Ann Arbor Science Publishers. Inc.. Ann Arbor. MI). Pittman. Russell W.. 1981. Issue in pollution control: Interplant cost dillerences and economies of scale. Land Economics 57. no. 1. l-17. Shephard, Ronald W.. 1970, Theory of cost and production functions (Princeton University Press, Princeton. NJ). Shephard, Ronald W, 1974. Indirect production functions, Mathematical Systems in Economics. no. 10 (Anton Hain. Meim&eim am Glad). Siegel, Sidney, 1956, Nonparametric statistics for the behavioral sciences (McGraw-Hill, New York). Thompson, Russell G, James A. Galloway and Lillian A. Navalanic, ads., 1977, The cost of electricity (Gulf Publishing Co, Houston, TX). Tran, Ngoc-Bich and V. Kerry Smith, 1983, The role of air and water residuals for steam electric power generation, Journal of Environmental Economics and Management 10, no. I. 3-9. U.S. Department of Ettargv. 1978. Steam-electric plant construction cost and annual production expensesz1975 (U.S.G~.O, Washington, DC).U.S. Federal Energy Regulatory Commission. 1979. Steam-electric plant air and water quality control data for the year ended December 31. 1975 (U.S.G.P.O., Washington, DC).