How green is my valley?

- evaluating the environment impact of products and production processes

by Jonathan Linton

Attempts have been made to compare the environmental impact or burden of different products or processes by equating different types of pollution. The results of these attempts are highly debated and time sensitive, due to the evolving understanding of the relative danger of different pollutants. In order to develop a method to better understand the magnitude of environmental impacts of a process or product, the relative impact on air pollution of electrical generation in forty-nine states is examined. Three pollutants (NOx, SOx, and CO2) are considered simultaneously. A States electrical generation is ranked as efficient if its environmental burden is relatively less then at least one other state in one or more of the three categories. States that have a greater environmental burden for air pollution in all of the three categories of air pollutants are ranked as inefficient and the degree of inefficiency is calculated. Data envelopment analysis (DEA) is used for this analysis and is able to reduce a set of forty-nine options to one efficient option. The limitation of DEA for analysis of environmental burdens is also discussed.


A method of ranking the environmental impact of products or processes that involve multiple incomparable pollutants is considered. Attempts at quantifying environmental damage, risk, and impact of the production of goods and services is ongoing. There are two reasons for continued activity in this area: (1) environmental degradation is a great concern to many people and (2) a way of comparing environmental impacts of alternative products or processes, that is widely accepted, is elusive. There have been many attempts to produce a simple environmental index that summarises incomparable environmental impacts with a single value but the presence of multiple incomparable impacts, and change in the relative importance of the impact as a factor of scientific knowledge (time) and location has made these attempts controversial and limited their acceptance.

It is possible that the complexity of comparing environmental impacts may prevent a widely acceptable single variable environmental index from ever being identified. In this case, the value of these single variable indexes may be limited to assisting a manufacturer in understanding whether the quantity of their production of emissions is increasing or decreasing (for example, see Nortel). It will not, however, indicate whether or not the organisations environmental impact or footprint is decreasing. One can argue that all other things being equal an overall decrease in quantity will result in a decrease in overall impact. Unfortunately, in environmental science it has been found too often that one pollutant can have an impact several orders of magnitude greater than another pollutant.

If we accept, at least for now, that a single environmental index is unrealistic, an alternative should be considered. Operations Research techniques have proven useful for assisting in decision making for a variety of environment related concerns (examples include: Salminen et al and Bloemhof-Ruwaard et al). However, operations research techniques have not been applied to obtain an understanding of the relative attractiveness of environmental impacts of products. The proposed method compares the generation of incomparable impacts to determine if one or more of the products or processes is clearly superior to its alternatives. In other words, which subset of a group of alternatives minimizes the emission of each of the incomparable pollutants, under consideration. By using such a method, we can avoid decisions that result in greater negative environmental impacts and a lower production of the desired output. Because this type of method does not place a relative value on incomparable impacts it will not identify the best product, but will identify the group that has the best overall impact given any weighting or conversion factor that could be applied to any of the environmental impacts under study.

The goal set out in the prior paragraph can be accomplished through the use of data envelopment analysis (DEA). The next section addresses DEA in general and more specifically the DEA model that is used for examining incomparable impacts. Finally, the use of DEA for examining environmental impacts is demonstrated by considering the emission of air pollutants by utilities that generate electricity in 49 US States.

The Data Envelopment Analysis Method

The purpose of this study is to evaluate a set of pollution generators, relative to each another, in order to identify which pollution generators are more efficient. In other words which pollution generators produce less pollution per unit of desired output. This will allow for best practices to be noted and environmentally rational choices to be made. The tool to be utilised for this purpose is the Data Envelopment Analysis (DEA) methodology. DEA was developed, by Charnes et al, to measure the relative efficiencies of a set of comparable decision making units. Applications have included schools, maintenance crews and innovation implementation.

In many application areas, the problem of developing a measure of performance may become complicated by at least two factors. First, the economic outputs and inputs may not have their prices clearly defined, a challenge presented by environmental impacts. Consequently, clearly established prices are not available and flexibility in the choice of these prices is needed. The second potential difficulty in evaluating performance may be the presence of non-economic factors. Referring again to environmental impacts, if one argues that the climate and geography of a region makes it more susceptible to certain types of impacts then weather should be considered and a case can be made for including this factor as a type of environmental input. The problem is then to determine how to weight such an input relative to other more tangible economic variables.

The basic concept in DEA is to measure the performance of a Decision Making Unit (DMU) against a projected point on an efficient frontier. In the application setting described here, states are viewed as decision making units. This paper demonstrates the use of DEA to determine which of a series of options (power production of US states) are environmentally efficient. Furthermore, DEA is used to produce an estimation of the percentage reduction in pollutants required to make an inefficient option, a states power production, efficient. To demonstrate the novel application of data envelopment analysis for comparing environmental impacts, data from the electrical generation industry is considered. The electrical industry is used as an example, because reliable data is publicly available. Unfortunately, environmental impact data is often unreliable or unavailable (see Linton). It is acknowledged that there are a large number of factors not considered in this paper such as the age of generating facility and the means of power generation (nuclear, hydro, coal, or other sources). However, the data on air pollution from power generation is suitable for a demonstration of the ability of Data Envelopment Analysis (DEA) to produce a subset of efficient options that are greener than the other, inefficient, set members no matter what the relative weightings of the pollutants are.

Example: Rating Polluting Efficiency of a Series of Options

Data envelopment analysis and other methods have been used previously to obtain insights into the efficiency of the electrical generation industry. However, these studies examine the relative efficiencies of generators based on a variety of input and output factors. This paper considers the efficiency of power generators at producing electricity while minimising the emission of air pollutants. Data on electrical power generation and associated air pollution is taken for forty-nine jurisdictions in the United States, see Table 1. (Vermont is eliminated from this analysis, due to the zero values for two of the three pollutants, due to the lack of precision of the data.) Three environmental impacts are considered carbon dioxide, nitrogen oxides, and sulphur dioxide. These impacts are a result of the generation of power in millions of kW hours. The first decision to be made for the DEA analysis is what constitutes an input and an output variable? Common sense may suggest that the three environmental impacts and power generation are all output variables. But we must consider that the environmental impacts are an unwanted by-product of the generation of power. Consequently, we wish to minimise the volume of the air pollutants for a unit of power generation.

Table 1: Electrical generation (in Million KiloWatt Hours) and air pollutants (in Thousand Short Tons) for fifty jurisdictions and the associated efficiency required for air pollution to be on the efficient frontier.

Not included
4,761 0 0 11
Oregon 1.0000 44,964 2 2 539
California 0.3559 132,264 15 149 38278
Alaska 0.1934 4,388 1 1 480
Washington 0.1060 87,121 63 41 9848
Maine 0.1020 11,624 15 5 2854
Connecticut 0.0580 34,277 66 26 13335
Montanna 0.0396 25,823 29 61 19051
South Carolina 0.0358 66,984 170 83 25190
New York 0.0341 130,454 426 169 73864
Utah 0.0302 30,496 44 85 30864
New Jersey 0.0288 41,144 90 69 17127
District of Columbia 0.0283 650 4 1 716
Louisiana 0.0264 54,292 91 123 37234
South Dakota 0.0262 6,985 30 19 3201
Illinois 0.0256 126,841 889 320 59487
Hawaii 0.0250 7,949 26 14 6892
North Carolina 0.0225 87,103 351 172 51312
Virginia 0.0224 43,360 189 86 29095
Arkansas 0.0221 33,400 67 73 23232
Mississippi 0.0208 21,055 117 45 12449
Oklahoma 0.0205 44,448 96 124 37150
Nebraska 0.0199 21,089 47 70 14031
Arizona 0.0192 53,125 133 123 37505
Wyoming 0.0192 36,751 85 137 41888
Georgia 0.0190 92,448 835 216 64870
Tenessee 0.0189 73,959 762 177 46851
Massachusetts 0.0179 39,158 267 97 31332
Alabama 0.0179 77,600 564 193 55300
Pennsylvania 0.0176 154,790 1266 390 118559
Texas 0.0174 232,380 595 617 198448
New Mexico 0.0174 28,349 72 89 31853
Florida 0.0171 124,254 712 322 94834
Minnesota 0.0168 40,377 152 107 33547
Maryland 0.0156 35,759 298 102 30981
Michigan 0.0150 91,518 404 278 73365
Colorado 0.0147 32,299 98 111 33184
Wisconsin 0.0143 44,361 290 144 37086
Nevada 0.0143 19,681 61 62 20238
Kansas 0.0142 34,234 169 107 29046
New Hampshire 0.0138 7,125 80 26 6194
Montana 0.0131 59,382 886 272 54191
Iowa 0.0127 28,080 189 98 28572
Ohio 0.0126 131,432 2370 536 124813
Delaware 0.0121 8,458 85 31 9494
Kentucky 0.0120 70,760 751 306 70594
West Virginia 0.0119 82,860 1007 325 83728
Rhode Island 0.0112 496 2 2 532
North Dakota 0.0112 25,705 160 102 28997
Indiana 0.0102 88,563 1564 475 104302

Minimisation of variables for a given quantity of another variable corresponds to an input oriented model, with power generation as the output and the environmental impacts as the inputs.

Consideration is also given to whether to treat the production of product using a Constant-Returns to Scale (CRS) model or Variable-Returns to Scale (VRS) model. If we assume that returns are constant to scale it suggests that the pollutant generated for each unit of output is constant. If VRS is assumed, it suggests that the marginal and average costs, in terms of air pollution, will decline with increasing output of product — in this case electricity. For this study, a Constant-Return to Scale (CRS) model is assumed.

Table 2: Electrical generation (in Million KiloWatt Hours) divided by air pollutants (in Thousand Short Tons) for fifty jurisdictions and the associated efficiency required for air pollution to be on the efficient frontier.

  Generation of Electricity/Pollutant
Efficiency (Iota)
Sulphur Oxides
Nitrogen Oxides
Carbon Dioxide
Vermont Not included Not Available Not Available 433
Oregon 1.0000 22,482 22,482 83
California 0.3559 8,818 888 3
Alaska 0.1934 4,388 4,388 9
Washington 0.1060 1,383 2,125 9
Maine 0.1020 775 2,325 4
Connecticut 0.0580 519 1,318 3
Montanna 0.0396 890 423 1
South Carolina 0.0358 394 807 3
New York 0.0341 306 772 2
Utah 0.0302 693 359 1
New Jersey 0.0288 457 596 2
District of Columbia 0.0283 163 650 1
Louisiana 0.0264 597 441 1
South Dakota 0.0262 233 368 2
Illinois 0.0256 143 396 2
Hawaii 0.0250 306 568 1
North Carolina 0.0225 248 506 2
Virginia 0.0224 229 504 1
Arkansas 0.0221 499 458 1
Mississippi 0.0208 180 468 2
Oklahoma 0.0205 463 358 1
Nebraska 0.0199 449 301 2
Arizona 0.0192 399 432 1
Wyoming 0.0192 432 268 1
Georgia 0.0190 111 428 1
Tenessee 0.0189 97 418 2
Massachusetts 0.0179 147 404 1
Alabama 0.0179 138 402 1
Pennsylvania 0.0176 122 397 1
Texas 0.0174 391 377 1
New Mexico 0.0174 394 319 1
Florida 0.0171 175 386 1
Minnesota 0.0168 266 377 1
Maryland 0.0156 120 351 1
Michigan 0.0150 227 329 1
Colorado 0.0147 330 291 1
Wisconsin 0.0143 153 308 1
Nevada 0.0143 323 317 1
Kansas 0.0142 203 320 1
New Hampshire 0.0138 89 274 1
Montana 0.0131 67 218 1
Iowa 0.0127 149 287 1
Ohio 0.0126 55 245 1
Delaware 0.0121 100 273 1
Kentucky 0.0120 94 231 1
West Virginia 0.0119 82 255 1
Rhode Island 0.0112 248 248 1
North Dakota 0.0112 161 252 1
Indiana 0.0102 57 186 1

The analysis of alternative pollution scenarios will now be demonstrated using DEA and will be discussed further in the implications and conclusions section. In the following section we discuss the use of DEA in the area of evaluating environmental impacts of products and processes. More specifically the results of a DEA analysis of electrical generation are considered.


An input oriented DEA model found (see Table 1) that most of the jurisdictions were inefficient. All inefficient jurisdictions are clearly dominated by the efficient jurisdiction — Oregon. In other words, lower levels of all three pollutants result from switching from power generation in an inefficient jurisdiction to Oregon’s power generation (see Table 2), all other things being equal. In Table 2, the output in power generation is shown divided by each of the respective inputs. For clarity, the ‘efficient’ generator of electricity is shown in a bold font. The efficient energy generator, that minimises pollution for each kWh produced, is Oregon. Energy use in any of the other states results in more air pollution than Oregon. For example, New York is an inefficient energy generator, for it to become an efficient generator would require a proportionate reduction to about 3.41% (Iota = 0.0341) of the current inputs (pollutants) while producing the same quantity of output. An examination of the data shows that after Oregon the most efficient states are: California, Alaska, Washington, and Maine with efficiencies of 0.3559, 0.1934, 0.1060, and 0.1020, respectively. All other jurisdictions have an iota of less than 0.1 — all of these states require a decrease of pollutants at least an order of magnitude to become efficient.

Implications and conclusions

It is unlikely that an individual will move from one state to another to reduce the amount of air pollution that is produced when they watch television or conduct any other activities that consumes electricity. Furthermore, it can be argued that this specific analysis is misleading since electricity generated in one state is frequently used in another state. However, this data clearly demonstrates how a series of products or processes with multiple incomparable environmental impacts can be compared to each other. And that in this process a large number of options can be reduced to a small set of options (a decrease in this case by well over an order of magnitude).

If a decision is made based on environmental impact forty-eight of forty-nine cases under study in the electrical generation example (in Tables 1 and 2) are rejected. Consequently, we can state that DEA is a powerful tool to rate the ‘greenness’ of either products or processes with outputs of comparable value. DEA offers a non-controversial method of obtaining a single metric that in some cases will offer sufficient information for decision making. DEA allows for the identification of manufacturing facilities that produce the smallest quantity of pollutant per product (or value added). Consequently, DEA can be used in clean production benchmarking studies for identifying the facilities with ‘best practices. Another use for DEA is to determine what product choice offers the most economic benefit for the lowest environmental impact.

Finally, by identifying the set of efficient decision making units (DMUs) we can determine for what range of coefficients there are two or more efficient DMU’s (not an issue with this data set). By examining what weights (for example, a factor to convert SOx impact to NOx impact equivalents) are necessary for different members of the efficient set to be preferable it is possible to reduce the number of efficient alternatives further — even to as few as one alternative. The elimination of DMU’s that are efficient requires agreement on what range of weightings or conversion factors is not acceptable. This process results in a clear indication of which efficient set member is preferable for which range of weights (conversion factors from one environmental impact to another). Consequently, it is easier to obtain agreement than the currently proposed single weighting systems that require agreement on a specific value for conversion factors.

Even though the relationships between product and impacts are equivalent to an input oriented model with pollutants as inputs and product as outputs, consideration of environmental impacts as inputs and a desired good or service as an output may make some uncomfortable. If pollution is considered not as an undesired by-product, but as a reduction in the planet’s carrying capacity for accepting additional units of pollutant then the specification of pollutants as inputs is consistent with traditional DEA applications. The earth can absorb a certain quantity of pollutant, by producing pollutants some of this carrying capacity is absorbed. Consequently, a unit of pollution emitted is a reduction, by a unit, of the planet’s carrying capacity for that pollutant. The carrying capacity perspective is consistent with the moral philosophy and economics literature — pollution may be considered as an input.

DEA offers a non-controversial option to compare controversial environmental metrics. In certain cases, DEA will offer sufficient information for decision making. However, if DEA does not offer sufficient information to reduce the number of possible efficient producers or products it still facilitates the choice of one or more ‘best solutions’.

This is accomplished by reducing the task of the analyst from choosing conversion factors or weights of incomparable environmental impacts to making decisions when presented with a range of possible conversion factors. For the decision-maker, DEA offers a widely accepted method to obtain non-controversial input to a decision that is frequently controversial.

For the interested reader

  • Kjeldgaard EA, Waste minimisation makes good business sense: The role of cost/risk/benefit analysis in project selection. International Journal of Environmentally Conscious Manufacturing(1): 75-84, 1992.
  • Graedel TE, Allenby, BR. Industrial Ecology. New Jersey: Prentice Hall, 1995.
  • Hertwich EG, Pease WS, Koshland CP, Evaluating the environmental impact of products and production processes - A comparison of six methods. Working paper. Berkeley: Department of Environmental Health Science, University of California, 1996.
  • Nortel. Northern Telecom Environmental Performance Index (EPI). Missisauga, Canada: Nortel, 1993.
  • Hart S, Beyond Greening, Harvard Business Review (1): 67-76, 1997.
  • Salminen P, Hokkanen J, Lahdelma R, Comparing Multicriteria Methods in the Context of Environmental Problems. European Journal of Operational Research 104(3): 485-496, 1998.
  • Bloemhof-Ruwaard JM, van Beck P, Hordijk L, Van Wassenhove LN, Interactions Between Operational Research and Environment Management. European Journal of Operational Research 85(2):229-243,1995.
  • Charnes AW, Cooper W, Rhodes E, Measuring the efficiency of decision making units. European Journal of Operational Research 2(6): 429-44, 1978.
  • Charnes AW, Cooper W, Rhodes E, Evaluating program and managerial efficiency: An application of data envelopment analysis to program follow through. Management Science 27(6): 668-97, 1981.
  • Cook W, Roll Y, Kazakov A, A DEA model for measuring the relative efficiency of highway maintenance patrols.INFOR 28(2): 113-20.
  • Linton JD, Cook WD, Technology Implementation: A comparative study of Canadian and US Factories. INFOR 36(3): 142-150, 1998.
  • Linton JD, An Investigation of the Feasibility of Using a Material Impact Database for Calculating the Impact of Engineering Designs. Kingston, Canada: Proceedings of the 6th International Seminar on Life-cycle Engineering, 432-442, 1995.
  • Goto M, Tsutsui M, Comparison of productive and cost efficiencies among Japanese and US electrical utilities. OMEGA: International Journal of Management Science 26: 177-194, 1998.
  • Fare, Grosskopf, Logan, The comparative efficiency of western coal-fired steam-electric generating plants: 1977-1979. Engineering Costs and Production Economics 11:21-30, 1987.
  • DOE (Department of Energy), Electric Power Annual 1989, Washington, DC: Energy Information Administration, Department of Energy, 1991.
  • Hardin G, Living within Limits, New York: Oxford University, 1993.
  • Meadows D L, The Limits to Growth; A Report for the Club of Rome’s Project on the Predicament of Mankind, New York: Universe Books, 1972.

JONATHAN LINTON is an Assistant Professor of Management at Polytechnic University of New York. He prefers to examine questions that overlap environmental, Operations and technology management. Dr Linton maintains close ties with industry through writing the occasional article on technology or environmental management for trade magazines. Email: jlinton [at]

First published to members of the Operational Research Society in OR Insight July- September 1999

Latest News

March 2024

Maths Summit to bring political leaders together with experts from academia and industry

On 12 March 2024, the Council for Mathematical Sciences and the ‘Protect Pure Maths’ project are putting on the Maths Summit, with the support The OR Society and four other learned mathematical societies. The intention is to bring together political leaders and experts from academia and industry to discuss how the mathematical sciences can contribute better to research, innovation and prosperity.

Read More

February 2024

Sophie Huiberts wins Gijs de Leve Prize

The Dutch national network for operational research, LNMB, has awarded the prestigious Gijs de Leve Prize to Sophie Huiberts, the award’s first female winner. The prize is given to the best PhD thesis in Mathematics of Operations Research over the period of consideration.

Read More

February 2024

Supply Chain Management Award won by Philip Morris International

This week it was announced that tobacco giant Philip Morris International (PMI) won the prestigious Supply Chain Management Award 2023 with their platform Sync Hub.

Read More