Application of Computers and Operations Research in the Mineral Industry –
Dessureault, Ganguli, Kecojevic & Dwyer (eds)
© 2005 Taylor & Francis Group, London, ISBN 04 1537 449 9
Technical efficiency analysis in Illinois coal mining sector
A.R. Samal
Department of Geology, Southern Illinois University at Carbondale, USA
M.K. Mohanty
Department of Mining and Mineral Resources Engineering, Southern Illinois University at Carbondale, USA
S.C. Sharma
Department of Economics, Southern Illinois University at Carbondale, USA
ABSTRACT: The annual coal production in a mine is a function of the input parameters, such as capital,
labor, technology, etc. utilized in the mining operation. Irrespective of the amount of coal produced, a mine can
operate technically efficiently if the production is maximized with optimal use of inputs with no quantitative
increase. The reason for an industry producing below the production frontier can be technically-inefficiency.
No raw-material is required to produce coal from mines. The inputs for producing coal are capital (invested on
machineries and setting up of infrastructure) and labor. This paper demonstrates a methodology to at analyze
efficiency of coal-mines using stochastic frontier approach. A translog production function is used along with a
combination of coal mining data of Illinois collected from the Key Stone Coal Manual and some data generated
based on discussions with mine owners.
1 INTRODUCTION a unit or the entire firm is the ratio of the output(s) to
the inputs.
Illinois is the fifth major coal producing state in the US.
With change in technology and economy, the number
of coal producing projects has decreased since early
1990s. The annual coal production in a mine is a func- The relation between the input (x) and output (y) are
tion of the inputs used. The inputs used to extract coal discussed by Coelli et al. (1998) as shown in figure 1.
from coal-mines are capital (invested on machineries For each unit of inputs the maximum output attainable
and setting up of infrastructure) and labor to produce can be derived from this production frontier. So, a tech-
minerals.A mine is said to be technically efficient if the nically efficient firm operates on this frontier. A firm
production is maximized with optimal use of inputs. A can not operate above the frontier. A firm operating
mine may produce below the production frontier due to below this frontier is said to be technically inefficient.
the technical-inefficiency or due to some other reasons The point A is an inefficient point, as a firm can attain
beyond the control of the management. To estimate the point B with no more input. The firm can get same
technical efficiency of the production activity and to output at point C with less input. The slope of the
identify the factors causing the technical inefficiency line joining the origin and point A can increase to
has been of interest to the economists and industry as reach the point B and then C. At C, the line becomes
this can lead to finding ways to maximize the technical tangent to the production frontier and the productiv-
efficiency to improve productivity. ity is at its maximum attainable level. So, a firm can
An industry uses some inputs to produce the output. be technically efficient but may not be at maximum
The inputs vary with the type of industry and with the productivity.
output too. The production function is presented as Keeping the inputs constant, the output may change
y=f (xi ), where y is the output produced and xi is the over time due to changes in technical limits of the
vector of inputs. The inputs for producing coal can be production process. So, as seen in figure 2, the tech-
labor and capital. The productivity ratio is one of the nical change from period-0 to period-1 would cause
many ways to measure the performance of a single in shift of production function from F0 ’ to F1 ’. In pro-
productive unit or units in a firm. Productivity of such duction industry, such as coal-mines, installation of a
49
Copyright © 2005 Taylor & Francis Group plc, London, UK
(2001). Kulshrestha and Parikh (2002) used the DEA
B approach and measured the technical efficiency of
Y
Indian coal mines. Zhang and Zhang (2001) used
stochastic production function to measure technical
C A efficiency of China’s large and medium sized iron and
steel industries.
This project aims at estimating technical efficiency
of coal-mines in Illinois State using stochastic frontier
approach. The translog production function is used in
this study. The labor and production data used in this
study is obtained from the Keystone Coal Industry
Manual. Due to unavailability of the capital data, we
derived the capital data based the production figures
X of individual mine and capital investment per ton coal
production. The capital derived in this way is not an
Figure 1. Production Frontiers and Technical efficiencies exact measure of capital in production. The findings of
(Coelli, Rao and Battese, 1998). this study do not reflect the exact estimation of tech-
nical efficiency of coal mines of Illinois. This study
F1’ is just an illustration of the use of stochastic frontier
y methodology to the mining industry.
F0’
2 STOCHASTIC FRONTIER METHODOLOGY
There are two major methods to measure relative tech-
nical efficiency among firms. The data envelopment
analysis (DEA) is non-parametric method and stochas-
tic frontier analysis (SFA) is a parametric method of
analysis. This study will use SFA of relative technical-
efficiency analysis of the coal mining projects in the
state of Illinois. Panel data is used to study the change
in efficiency over a time period.
x
0
2.1 Stochastic frontier analysis:
Figure 2. Production function with technical-change in two
periods. According to Farrell (1957), frontier production func-
tion is estimated from using a parametric function
using data of N firms. The model is given by
new machine or, replacement of an old machine with
a new machine or, change of production procedure etc
are examples of technical changes that would cause
shift of production function. where, yi is the observed output, and xi is the vec-
The efficiency of coal producing industries in Illi- tor of inputs, β = (β0 , β1 , β2 , β3 , ……….., βk ) is
nois is studied by Byrnes et al (1984), Thompson a (k + 1) vector of unknown parameters and ui is
et al (1995). Byrnes, Fare and Grosskopf (BFG, 1984) the non-negative random variable associated with the
focused on the measurement of technical efficiency technical inefficiency in the production of firms (Bera
using linear programming. In the DEA approach (Data and Sharma, 1999). This is in the Cobb-Douglas form
Envelopment Analysis) of Thompson, Dharmapala of production function and proposed by Aigner and
andThrall (TDT, 1995) to measure technical efficiency Chu (1968).
of Illinois coal mines the data set of BFG model is used. From equation 2, we derive
The data is reformulated to derive three major inputs
for the only output (thousand tons of coal mined).
The input data were: labor (thousands of labor miner
day), capital (106 dollars spent on instruments) and
mine quality (tons of clean coal per thousand tons of
over-burden).
Other notable research works in this line are that of
Kulshrestha and Parikh (2002) and Zhang and Zhang
50
Copyright © 2005 Taylor & Francis Group plc, London, UK
The observed output, yi , is bound above by the Table 1. Parameter estimates of the translog function.
non-stochastic quantity, exp (xi β), making it a deter-
ministic frontier model. This model does not account Standard
for influence of measurement errors and other noise. Variable Parameter Estimates error
To accommodate the effects of noise, the stochastic
Constant β0 4.4998 1.4046
frontier approach has been developed. The foremost
ln Lit β1 −1.7699 0.2868
researchers proposing the stochastic frontier approach ln Kit β2 0.9497 0.1278
are Aigner, Lovell (1977) and Schmidt and Meeusen t β3 −0.0081 0.0293
and Broeck (1977) and later Battese and Corra (1977) 0.5(ln Kit )2 β4 −0.0068 0.0049
as discussed in Coelli, Rao and Battese, (1998). A 0.5(ln Lit )2 β5 0.2215 0.0496
composed error structure has been proposed by them. t2 β6 −0.0003 0.0012
The composed errors has two error terms; an addi- (ln Kit )*(ln Lit ) β7 0.0239 0.0159
tional random error term vi and the non negative error t*ln Kit β8 −0.0013 0.0022
term ui . t*ln Lit β9 0.0096 0.0044
2 2 2
σ = σu + σv 0.0065 0.0011
γ = σu2 /(σ u2 + σv2 ) 0.0832 0.1562
µ 0.0465 0.0553
η 0.0223 0.0625
The random error, vi is due to statistical errors, log likelihood function = 130.1791
effects of weather, strikes, luck, unexpected accidents
etc. The vi s can be positive or negative (Coelli, Rao
and Battese, 1998, pp 186). The production unit can
be above or, below the production frontier depending mine at time t and Lit is the labor of ith mine at time t.
on the enormity of vi . The vi s are independent of ui s. β0 to β9 are unknown parameters to be estimated.
In a time-varying model (panel data), the ineffi- The technical efficiency (TEit ) of any firm (i) at any
ciency effect (uit ) of any firm (i) at any time period (t) time (t) = exp(−uit ).
is ui {exp[−η(t − T)]}, where i = 1, 2, 3, 4, …….., N;
t = 1, 2, 3, 4, ………, T and η is an unknown parameter
to be estimated. If t = T, uit ≡ ui . So the random vari-
able ui can be considered as the technical inefficiency For a completely efficient firm, TEit = 1, which is
effect of the ith firm in the last period of the panel. impractical. The TEit goes to 0 for a completely
If η < 0, uit ≥ ui . If η < 0, uit ≤ ui . A special case of inefficient firm.
time varying inefficiency model is the time-invariant
model, when η = 0. Importance of η in time-varying
inefficiency model is discussed by Coelli, Rao and 3 DATA ANALYSIS AND RESULTS
Battese (1998). Given the η value, it is easy to com-
pare the technical inefficiency effect (uit ) of the firms The dataset used in this project is gathered from major
over a time period. coal industry data publication, Keystone Coal Manual
(1989–2001) and energy information administration
2.2 Estimation of production function: (e.i.a.) and other sources such as ICCI (Illinois Clean
In the stochastic frontier approach, the translog func- Coal Institute). In 1989 a total of 38 coal mining units
tion is commonly used to estimate technical efficiency operated in Illinois State. These included 15 surface-
(Aigner & Lovell, 1977, Schmidt, Meeusen & Broeck, mines and rest are underground mines. The number
1977, Battese and Corra, 1977), Coelli, Rao and of coal mining units reduced to 18 in 2001, out of
Battese, 1998). which six were surface-mine projects are producing
Using Battese and Coelli (1992), the translog pro- coal in addition to some projects started after 1989.
duction function of a panel data will have the form Out of these 15 projects, data for some projects are
(Sharma, Sylwester, Margono, 2003): not available. Out of all these only nine (09) projects
are chosen (table 2). Except the mine D, which is a
surface-mine, all mines are underground mines. Here
the data related to these years for which all these units
produced coal are considered. The data include output
(tons of coal, Y), labor (employees in each project, L)
and capital (K) spent on different machinery used in
mining, which does not include coal preparation.
where i = 1, 2, 3,……… denotes the number of mines Due to unavailability of the capital data, data on the
and t = 1, 2, 3 …. denotes the time in years. Where Yit capital input is calculated from using the capital-
is the output of ith mine at time t, Kit is the capital of ith investment per ton coal production data, which was
51
Copyright © 2005 Taylor & Francis Group plc, London, UK
Table 2. Efficiency measures of Illinois coal mines.
Mine
Mean
Year A B C D E F G H I efficiency
1989 0.9429 0.9275 0.9445 0.9804 0.9018 0.9473 0.9197 0.9473 0.9257 0.9375
1990 0.9441 0.9291 0.9457 0.9809 0.9039 0.9485 0.9214 0.9485 0.9273 0.9388
1991 0.9454 0.9306 0.9469 0.9813 0.906 0.9496 0.9231 0.9496 0.9289 0.9402
1992 0.9466 0.9321 0.9481 0.9817 0.908 0.9507 0.9248 0.9507 0.9304 0.9415
1993 0.9477 0.9336 0.9492 0.9821 0.91 0.9518 0.9264 0.9518 0.932 0.9427
1994 0.9489 0.9351 0.9504 0.9825 0.912 0.9528 0.928 0.9529 0.9334 0.944
1995 0.95 0.9365 0.9515 0.9829 0.9139 0.9539 0.9296 0.9539 0.9349 0.9452
1996 0.9511 0.9379 0.9525 0.9833 0.9157 0.9549 0.9311 0.9549 0.9363 0.9464
1997 0.9522 0.9392 0.9536 0.9837 0.9175 0.9559 0.9326 0.9559 0.9377 0.9476
1998 0.9532 0.9406 0.9546 0.984 0.9193 0.9569 0.9341 0.9569 0.9391 0.9487
1999 0.9543 0.9419 0.9556 0.9844 0.9211 0.9578 0.9355 0.9578 0.9404 0.9499
2000 0.9553 0.9431 0.9566 0.9847 0.9228 0.9587 0.9369 0.9588 0.9417 0.951
2001 0.9563 0.9444 0.9575 0.9851 0.9245 0.9597 0.9383 0.9597 0.943 0.952
Efficiency of Illinois coal mines
(η = 0.0229)
0.99 Mine A
0.98 Mine B
0.97
Mine C
0.96
Mine D
Efficiency
0.95
Mine E
0.94
Mine F
0.93
Mine G
0.92
Mine H
0.91
Mine I
0.90
1 2 3 4 5 6 7 8 9 10 11 12 13
Years
Figure 3. Efficiency measures of Illinois coal mines.
collected by personal communication with mine- So, per year capital = total capital cost * CRF
owners of Illinois. For surface-mines with stripping where, total capital cost = tons of coal mined * capital
ratio 10:1, the capital investment is estimated to be investment per ton of coal. Capital were converted to
$15.00 per ton of coal produced, where as for stripping 2001 USD values using conversion factor. The translog
ratio of 20:1, the capital investment would be $20.00. production function used here is given by equation
For underground coal production, $25.00 is spent for 6 and discussed in detailed by Sharma, Sylwester,
producing a ton of coal. Assuming average life of mine Margono, 2003. Parameter estimates are reported in
20 years and a discount rate of 12%, the capital recov- table 1. The technical efficiency estimates of each firm
ery factor (CRF) of 0.1339 is used to calculate annual (here mining units) of all years and mean technical
capital investment. efficiency of that year are given in Table 2.
52
Copyright © 2005 Taylor & Francis Group plc, London, UK
4 INTERPRETATION AND DISCUSSION application of the software FRONTIER version 4.1.
Also I acknowledge the help of Mr Vishal Gupta for
The perfectly efficient mining project will have a his cooperation.
technical efficiency value of 1.0, which is practi-
cally unattainable. Perfectly inefficient project will
have technical efficiency factor ‘0’. In this project REFERENCES
no project is detected as fully efficient, and none is a
perfectly inefficient project. The technical efficiency Bera, A.K. and Sharma, S.C. (1999) “Estimating Production
factors are plotted against time (t, in year) to compare Uncertainty in Stochastic Frontier Function Models”
the technical efficiency of different mining projects Battese, G.E. and Coelli, T.J. (1992) “Frontier Production
Functions, Technical Efficiency and Panel Data: With
(Figure 3).
Application to Paddy Farmers in India”, Journal of Pro-
As seen in figure 3, the η value being positive, the ductivity Analysis, 3, 153–169
technical efficiency increases for all projects in these Boyd, G.A. (1984) “Scale and Productivity in surface Coal
13 years. The mine D (Surface-mine) project seems Mining”, PhD Dissertation, SIU-Carbondale
to be the most efficient mine followed by mine H, Byrnes, P. and Fare, R. (1987) “Surface Mining of Coal:
mine C and mine A mining projects. Excluding the Efficiency of U.S. Interior Mines”, Applied Economics
only surface-mining project (mine D), the other three 19(December):1665–73
projects listed above seem to have steady but slow Byrnes, P., Fare, R. and Grosskopf. S. (1984) “Measuring
increase in technical efficiency with time. The mine Productive Efficiency. An Application to Illinois Strip
Mines”, Management Science,Vol. 30, No. 6, pp. 671–681
E project of Coulterville, started with very low tech-
Coelli, T. (1996) “A Guide to FRONTIER Version 4.1: A
nical efficiency in 1989, but seems to have highest Computer Program for Stochastic Frontier Production and
rate of increasing technical efficiency in this period Cost Function Estimation”, CEPA Working Paper 96/08,
(13 years). University of New England
The γ [σu2 / (σ u2 + σv2 )] value varies between 0 and Coeli, T., Rao, D.S. Battese, G.E. (1998) “An Introduction to
1: a value nearer to one may be interpreted as the Efficiency and Productivity Analysis”. Kluwar Academic:
stochastic model not significantly different from deter- Boston
ministic model and significantly less random error Energy InformationAdministration, (2003) www.eia.doe.gov
(Coelli, Rao, Battese, 1998, pp 215). In this project, the Farrell, M.J. (1957) “The Measurement of Productive Effi-
ciency”, the Journal of the Royal Statistical Society A,
γ value is nearer to zero (0.0832 and standard error of
Vol. 120, pp. 253–281
0.1562). This result shows a presence of considerable Keystone Coal Industry Manual, (1991 to 2003) New York,
amount of random error (vit ) in the error term (εit ). McGraw-Hill
Kulashrestha, M. and Parikh, J.K. (2002) “A study on effi-
ciency and productivity of Coal Mining in the Indian Coal
5 CONCLUSION Sector”, Energy Economics 24, pp. 439–453
Kulashrestha, M. and Parikh, J.K. (2001) “A Study of Pro-
Coal mining is highly cost intensive activity. For bet- ductivity in the Indian Coal Sector”, Energy Policy, 29,
ter production, more and more mechanization of the 701713, IGIDR RP No. 506, 2001
mining process is seen in recent years. This process Myers, G.J., Fare, R., Yoon, B.J. (1981) “Productivity in strip
reduces human resource employment and increases mining of coal”, USDOE/CMTC Special Research
the capital investment for a higher production in more Shar, R. (2002?) “inflation conversion factors for dollars 1665
to estimated 2013” Http://oregonstate.edu/dept/pol_sci/
efficient way. In the chosen Illinois coal mines, the fac/sahr/sahr.htm#_download_conversion_factors_1
technical efficiency of production is mainly attributed Sharma, S.C., Sylwester, K., Margono, H. (2003) “Technical
to the mechanization of the mines and adaptation of Efficiency and Total factor Productivity analysis across
newer technology. U.S. States: 1977–2000” (Un-published paper)
Further analysis of all producing coal mines in Thompson, R.G., Dharmapala, P.S., Thrall, R.M. (1994)
Illinois with more detailed inputs may be helpful in “Linked-Cone DEA Profit Ratios and Technology Effi-
identification of the technically efficient projects and ciency with Application to Illinois Coal Mines”, Working
the factors responsible for the same. Taking scale- Paper No. 110, Rice University, pp. 1–22
economics into consideration, the analysis of all coal Zhang, X.G, Zhang, S., (2001) “Technical Efficiency in
China’s Iron and Steel Industry: evidence from the new
mines in Illinois can be an important facet of future census data”, International Review ofApplied Economics,
research. Other factors to consider are mechanization, Vol. 15, No. 2, pp. 199–211
skilled labor employment, coal quality.
ACKNOWLEDGEMENTS
My thanks go to Mr Heru Margono (Graduate Stu-
dent, SIU-C) for his cooperation in learning the
53
Copyright © 2005 Taylor & Francis Group plc, London, UK
