You are on page 1of 11

Journal of Cleaner Production 9 (2001) 417–427

www.cleanerproduction.net

Using data envelopment analysis to evaluate environmentally


conscious waste treatment technology
a,* b
Joseph Sarkis , Jeffrey Weinrach
a
Graduate School of Management, Clark University, 950 Main Street, Worcester, MA 01610-1477, USA
b
JCS Novation Inc., 4501 Indian School Rd. NE, Suite 105, Albuquerque, NM 87110, USA

Received 8 February 2000; accepted 29 November 2000

Abstract

As environmental issues become increasingly important for organizational operations, the roles of organizational practices and
technology in improving environmental performance increase. Some of these new practices and technologies are difficult to evaluate
using traditional approaches. In this paper we analyze a decision making case study concerning the investment and adoption of
environmentally conscious waste treatment technology in a government supported agency. Data envelopment analysis, and some
of its extensions, will be used to evaluate the technologies. The results show that this tool could be useful in evaluating these
technologies. Managerial insights on the use of these models, including caveats on the use of DEA, are also presented.  2001
Elsevier Science Ltd. All rights reserved.

Keywords: Technology evaluation; Data envelopment analysis

1. Introduction the steel and iron industry [2], telecommunications [3]


and universities [4], to name a few. A thorough review
Managing any organization requires the capability to of DEA theory and application can be found in [5].
effectively measure and analyze information that is gen- The use of DEA for evaluating intra-organizational
erated by business processes. Environmental manage- environmental decisions is the focus of this paper. An
ment of organizations and managing environmentally evaluation of environmentally conscious waste treatment
focused business decisions from the perspective of the technology for a governmentally supported institution is
natural environment is even more critical. These mana- the specific application introduced. The background of
gerial practices may be reactive (e.g. responding to the decision environment provides some of the issues
environmental regulation) or proactive, (e.g. competitive associated with the organizational environment and
reasons and for the satisfaction of the stakeholders). DEA’s use. We have also identified practical issues asso-
Developing and applying tools that have not been tra- ciated with DEA use that managers and researchers
ditionally utilized for environmental management at stra- should be aware of. These issues and a comparison of
tegic and operational levels could aid the managerial DEA (and its variations) with the return on investment
decision making processes. One tool that has gained (ROI) criterion, currently in use for evaluating the tech-
wide acceptance at an industrial level of analysis (e.g. nologies, is also presented. A general description of how
organizational comparisons) is the use of data envelop- DEA performs with respect to other multiple criteria
ment analysis (DEA). For example, DEA has been used evaluation tools are discussed. The paper concludes with
to evaluate the relative performance of a broad range of a summary of the paper and findings along with
public and private organizations including banking [1], additional issues for practice and research consideration.

2. The decision environment


* Corresponding author. Tel.: +1-508-793-7659; fax: +1-508-793-
8822. The Department of Energy is comprised of production
E-mail address: jsarkis@clarku.edu (J. Sarkis). plants and laboratories whose primary missions have

0959-6526/01/$ - see front matter  2001 Elsevier Science Ltd. All rights reserved.
PII: S 0 9 5 9 - 6 5 2 6 ( 0 0 ) 0 0 0 8 4 - 6
418 J. Sarkis, J. Weinrach / Journal of Cleaner Production 9 (2001) 417–427

been to develop and maintain the nuclear weapons stock- ides from Waste Salts-1’s goal is to ultimately have
pile for the Department of Defense. These activities have waste concentrations (outputs) at the transuranic lev-
resulted in a forty-year legacy of radioactive and mixed els, while Distillation and Separation of Actinides
(a combination of radioactive and hazardous) waste from Waste Salts-2’s goal is to have waste concen-
which must be managed. In order to prevent future waste trations at the low level waste level.
problems, a comprehensive pollution prevention pro- 4) Waste Acceptance for Nonradioactive Disposal is
gram has been adopted and is currently being a method to separate radioactive from non-radioactive
implemented at these sites. components of a solid waste stream using state-of-
As an integral part of any successful pollution preven- the-art sensor technology. Much of the “radioactive”
tion program, goals must be set and goals must be met. waste that ends up in a low-level waste landfill is
Since the Department of Energy consists of so many dif- actually not radioactive but “suspect” radioactive
ferent types of facilities and operations, these goals which means that it cannot be definitively determined
would naturally not be identical. However, there should that this waste is radioactive but since it came from
be (and needs to be) methods to compare the relative a site which historically has radioactive materials, it
success of meeting pollution prevention goals regardless is assumed to be radioactive. Waste Acceptance for
of what the goals are or what type of facility or operation Nonradioactive Disposal allows personnel to separate
is being addressed. DEA is one of the methods that can the radioactive components of waste from the non-
help address the problems that the Department of Energy radioactive components and, therefore, greatly reduce
faces in realizing its pollution prevention goals. the volume (and the cost) of waste management.
The plutonium facility at Los Alamos National Lab- 5) Electrolytic Decontamination of Gloveboxes is an
oratory, known as TA-55, generates a significant portion improved technique to decontaminate gloveboxes
of the radioactive and mixed waste at Los Alamos. In (metallic boxes that hold contaminated gloves and
order to implement successful pollution prevention pro- other materials) from transuranic waste to low-level
grams at TA-55, a variety of approaches are being taken. waste that, if successful, would greatly reduce the vol-
The eight technologies and new processes that are being ume of liquid and solid wastes generated during
used for this DEA study represent some of the process decontamination.
modifications that TA-55 personnel are currently 6) Chloride Aqueous Waste Stream Polishing deals
developing and implementing. Each new process either with the other significant process waste stream at TA-
addresses a radioactive mixed solid waste stream or a 55 (the nitrate stream is the other). The polishing pro-
radioactive mixed aqueous waste stream (in a few cess would make the neutralization of the chloride
cases, both). solution and actinide removal much easier and less
Here is a brief overview of the eight technologies costly through the use of new resins and adsorbents.
(processes) being evaluated in this study: 7) High-Gradient Magnetic Separation is a technique
to separate plutonium from other metals and solids
1) Nitric Acid Recovery and Recycle addresses the using a magnetic field. The paramagnetic plutonium
need to recover the acidic aqueous waste stream that wastes can be separated from most of the other
is generated during the nitrate aqueous plutonium pro- materials that are diamagnetic. In a similar way to
cessing stage. Waste Acceptance for Nonradioactive Disposal, this
2 and 3) Distillation (concentration) and Separation process would greatly reduce the amount of poten-
of Actinides from Waste Salts addresses the need to tially non-radioactive waste that ends up at a low-
separate much of the plutonium from the bulk waste level landfill.
salts that can be classified as transuranic waste which 8) Application of Freeze-Drying Technology to
would eventually be discarded at the Waste Isolation Decontamination of Liquid Wastes uses freeze-drying
Pilot Plant in Carlsbad, New Mexico. Plutonium sep- techniques to remove the acid and aqueous compo-
aration would classify the resultant solid waste nents of the waste streams from the radioactive con-
streams as Low-Level waste (based primarily on taminants.
concentrations) and therefore could be discarded in a
much less costly low-level waste landfill. There are Even though these technologies may be quite different
two alternatives that have been identified for this tech- in their approaches, they do have a few management
nology; we have defined them as Distillation and Sep- (environmental and general) characteristics in common:
aration of Actinides from Waste Salts-1 and Distil-
lation and Separation of Actinides from Waste Salts- 앫 They all address significant waste issues and would,
2. The differences focus on the concentrations and if implemented, greatly reduce health risks and costs.
processing needed to alter the wastes to either the 앫 They all require some technological advance; that is,
transuranic waste level or the low level of concen- they could not be implemented through administrat-
trations. That is, Distillation and Separation of Actin- ive controls.
J. Sarkis, J. Weinrach / Journal of Cleaner Production 9 (2001) 417–427 419

앫 They all require some time to develop, test, and initial efficiencies of the various systems under consider-
eventually implement and, therefore, need to be ation. The next phase will carry out a cross efficiency
addressed in stages. analysis using “arbitrary”, or simple, and aggressive
cross efficiency measurement approaches. The cross-
efficiency and ranking efficiency scores will then be
compared to the return-on-investment (ROI) scores from
3. Data calculations provided by the Los Alamos Laboratories
[6].
The information used for this paper is used for
illustrative purposes and is adapted from actual reports
for the evaluation of these technologies. Additional
applications and more robust data may provide 5. DEA, cross efficiency, and ranking models
additional insights.
To be able to use the data for the DEA evaluation In this section we provide a review of basic DEA and
models, detailed in the next section, inputs and outputs “cross-efficiency” extensions to the DEA models.
must be identified. The inputs are traditional cost meas- Productivity models have traditionally been used to
ures that include operating and capital investment costs measure efficiency of systems. Typically, DEA pro-
(these are also used to calculate ROI values and are ductivity models for a given “decision making unit” use
initially represented as cost savings). Even though the ratios based on the amount of output per given input. In
costs may be integrated into one value we felt that this paper a decision making unit will be a technology
allowing for their separation provided managerial under consideration within the organization. DEA allows
decision making flexibility (e.g. some management for the simultaneous analysis of multiple inputs to mul-
decision making may focus more heavily on capital tiple outputs, a multi-factor productivity approach. The
investment costs than operational costs). The output data general efficiency measure that is used by DEA can best
will be estimates of wastes generated with the introduc- be summarized by Eq. (1).
tion of the technology. The wastes will include low level
and transuranic wastes measured in terms of drums. Fig.
1 shows the input and output relationships that will be 冘 Osyvky
Eks⫽
y
inputs for the DEA models. (1)
There are eight technologies that are to be evaluated
in this model. For the basic DEA models, the number 冘x
Isxukx
technologies (or units under consideration) should be a
minimum of between 2 to 5 times the total number of where: (Eks) is the efficiency or productivity measure of
input and output factors. Thus, we have limited the total technology s, using the weights of “test” technology k;
number of inputs and outputs to four factors. The actual (Osy) is the value of output y for technology s; (Isx) is
values used in the data set for the problem we are the value for input x of technology s; (vky) is the weight
addressing will be described after discussion of the DEA assigned to technology k for output y; and (ukx) is the
models that will be used in this paper. weight assigned to technology k for input x.
For the traditional DEA ratio model developed by
Charnes et al. (CCR) [7] the objective is to maximize
4. Models and methodology the efficiency value of a test technology k, from among
a reference set of technologies s, by selecting the optimal
The analysis of technology evaluation will consist of weights associated with the input and output measures.
two major phases. We will use the standard ratio model The maximum efficiencies are constrained to 1. The for-
of DEA and a Ranking DEA model to determine the mulation is represented in Eq. (2).

Fig. 1. A graphical representation of the inputs and outputs and decision units (new technologies) for the DEA model.
420 J. Sarkis, J. Weinrach / Journal of Cleaner Production 9 (2001) 417–427

冘 Okyvky
ukx,vkyⱖ0 (4)

maximize Ekk⫽
y


Eq. (4), which we will call the rank efficiency formu-
Ikxukx lation, allows for technically efficient scores to be greater
x than 1. This result, according to Andersen and Petersen,
subject to: will allow for a more discriminating set of scores for
technically efficient units.
Eksⱕ1 for all technologies s One of the difficulties with simple efficiency (the CCR
ukx,vkyⱖ0 (2) and rank efficiency models) is a resulting set of “false
positives.” A false positive is a unit that may be efficient
under CCR, but its cross-efficiency score (described
This non-linear programming Eq. (2) is equivalent to below) is less than a CCR inefficient unit. A false posi-
the following linear programming Eq. (3) (see [7]): tive technology typically has the characteristic of weigh-


ing heavily on a single input or output, thus making it
maximize Ekk⫽ Okyvky more efficient than any other technology because of a
y dominant performance on a single factor. As mentioned,
subject to: a procedure for discriminating between “true” efficient
technologies and false positive technologies is by ana-
Eksⱕ1 for all technologies s lyzing the cross-efficiencies. Sexton, Silkman, and


Hogan [9], introduced the concept of cross-efficiencies
Ikxukx⫽1 and the cross efficiency matrix. The cross efficiency
x matrix provides information on the efficiency of a spe-
cific technology with the optimal weighting schemes
ukx,vkyⱖ0 (3)
determined for other technologies. Table 1 summarizes
a typical cross efficiency matrix. The kth row and the sth
The transformation is completed by constraining the column represent the efficiency measure of technology s
efficiency ratio denominator from Eq. (2) to a value of by the optimal weights for technology k (Eks.). Each of
the columns of the cross efficiency matrix may then be
1. This is represented by the constraint SIkxukx=1.
x averaged to get a mean cross efficiency measure (this
The result of Eq. (3) is an optimal “simple” or techni- calculation will be defined as the simple cross efficiency
cal efficiency value (Ekk*) that is at most equal to 1. If measure when the optimal weights that are being used
Ekk*=1, then it means that no other technology is more are from the basic CCR model) for each technology (es).
efficient than technology k for its selected weights. That A false positive technology is associated with an
is, Ekk*=1 has technology k on the optimal frontier and efficient technology that has a relatively small cross
is not dominated by any other technology. If Ekk*⬍1 efficiency value or a cross efficiency value that is less
then technology k does not lie on the optimal frontier than the cross efficiency value of an initially inef-
and there is at least one other technology that is more ficient technology.
efficient for the optimal set of weights determined by A pitfall in determining a cross efficiency score is that
Eq. (3). The Eq. (3) is executed s times, once for each the weights, derived from the CCR model, used to calcu-
technology. The first method in the analysis will use the late the optimal simple efficiencies (and eventually used
CCR model to calculate the simple efficiency. in cross efficiency measures) may not be unique. To
A variation of the CCR model proposed by Andersen overcome this difficulty, a formulation developed by
and Petersen [8], allows for the use of DEA efficiency Doyle and Green [10] — one that will help generate a
scores for ranking alternative units. In their model, they less ambiguous set of weights — may be used for cross
simply eliminate the test unit from the constraint set. The efficiency calculation and development of a cross
new formulation is represented by Eq. (4). efficiency matrix. This formulation is represented by
Eq. (5).

冘冉 冘 冊
maximize Ekk⫽ Okyvky
y minimize vky Osy ,
subject to: y s⫽k

Eksⱕ1 for all technologies s⫽k subject to:

冘x
Ikxukx⫽1 冘冉 冘 冊
x
ukx
s⫽k
Isx ⫽1,
J. Sarkis, J. Weinrach / Journal of Cleaner Production 9 (2001) 417–427 421

Table 1
Generalized cross efficiency matrix

Mean cross efficiencies scores


Rating technology 1 2 Rated technology 3 ... n

1 E11 E12 E13 ... E1n


2 E21 E22 E23 ... E2n
3 E31 E32 E33 ... E3n
: : : : :
: : : : :
n En1 En2 En3 ... Enn
e1 e2 e3 en

冘y
Okyvky⫺Ekk 冘 x
Ikxukx⫽0
nology to get a mean aggressive cross-efficiency
value.

Eksⱕ1 for all technology s⫽k


ukx,vkyⱖ0 (5)
6. Data scaling and initial results

The Doyle and Green formulation as presented in Eq. In the initial runs of the model we considered the dif-
(5) has a primary goal of obtaining a maximum simple ferential rates for the input and output data. That is, the
efficiency score for technology k (the test unit), and a marginal impact of the technology on costs, transuranic
secondary goal of determining a set of weights that will and low level wastes were used as the basic data points.
minimize the other technologies aggregate output. This Table 2 shows the data as initially captured. The first
model has been defined as an “aggressive” formulation. column identifies the technology. The second and third
To make the Doyle and Green formulation a “benevol- columns, respectively, represent the capital and oper-
ent” formulation — one where the secondary goal is to ational cost savings of the technology. Positive cost sav-
maximize the other technologies aggregate output — ings represent how much can actually be saved by a new
requires changing the “minimum” to “maximum” in technology implementation. A negative value represents
Eq. (5). an increase in costs. The final two columns represent the
Similar to the rank efficiency scores, the mean cross- amount of waste reduction in drums (a negative number
efficiency scores can be used to rank the technology in these columns represents an increase in that waste
alternatives. output).
In summary the following DEA-based performance DEA requires the data sets to be non-negative for the
measures will be calculated for each technology: outputs and strictly positive for the inputs. It is also
assumes that input values are improving as they get
앫 Simple Efficiency — is the efficiency score calculated smaller and output values are improving as they get
by the basic CCR model for each technology. larger. Thus, this data needs to be scaled for two reasons.
앫 Ranked Efficiency — is the efficiency score calcu- The first reason is to eliminate any negative values in
lated by the rank efficiency model for each tech- the data set. The next is to make sure that data set has
nology. the characteristic of having smaller values preferable for
앫 Simple Cross-Efficiency — is the efficiency score cal- inputs and larger values preferable for outputs. The
culated for a technology s by multiplying the optimal initial scaling required the cost categories (inputs) to be
weights of a technology k from the CCR model by altered such that lower values were preferable to higher
the original ratio of inputs and outputs of the tech- values. There was also a need to eliminate any negative
nology s. The cross-efficiency scores are then aver- numbers in any of the columns. The outputs (last two
aged for each technology to get a mean simple columns of Table 2) already have the increasing-value-
cross-efficiency. is-better characteristic.
앫 Aggressive Cross-Efficiency — is the efficiency score The scaling was initially accomplished by subtracting
calculated for a technology s by multiplying the opti- the values in the input columns from an arbitrarily large
mal weights of a technology k from Doyle and number (10,000 in this case). The output columns were
Green’s aggressive formulation to the original ratio scaled by adding a value of 10 to the transuranic output
of inputs and outputs of the technology s. The cross factor and a value of 20 to the low level waste output
efficiency scores are then averaged for each tech- factor. This scaling allowed each of the inputs and out-
422 J. Sarkis, J. Weinrach / Journal of Cleaner Production 9 (2001) 417–427

Table 2
Unscaled data set of results for input and output factors after introduction of new technology

Technology Inputs Outputs


Capital Cost Saving Operational Cost Savings Transuranic Waste Low Level Waste
(000’s) (000’s) Decrease (Drums) Decrease (Drums)

Nitric acid recovery 240 128 2 0


Distill and separate actinides/salts 1 ⫺325 1006 0 59
Distill and separate actinides/salts 2 ⫺1400 1017 27 26
Waste accept. for non-rad. disp. ⫺363 500 0 75
Decon. gloveboxes ⫺55 256 77 66
Chloride waste stream polishing 4450 ⫺246 12 ⫺14
Magnetic separation 6520 79 ⫺3 ⫺14
Freeze drying technology 3900 341 20 23

puts to have the characteristics to run the DEA model. play a role in the final results of this analysis. First,
These values are shown in Table 3. An initial execution instead of using arbitrary numbers, actual values were
of the data using the CCR model of DEA provided us used for the initial base line values. This first step helped
the results as shown in the final column of Table 3. To eliminate any negative values from the data set. The
determine if there was any sensitivity from the arbitrary actual baselines were 7500, 1500, 280, and 1400 for
additive scaling procedure used, we evaluated another capital costs, operational costs, TRU and LLW respect-
set of scaling values. This additional test scaling ively. The values within each column were subtracted
included subtracting each element in the capital cost from their respective baseline. These results provided
input and operational costs input from values of 7000 appropriate characteristics for the inputs, but the outputs
and 1500, respectively. Values of 4 and 15 were added (TRU and LLW) still needed to have larger values as
to the transuranic and low level waste columns, respect- more preferable. These columns were scaled by taking
ively. The new results are shown in Table 4. The the inverse of their actual values, thus maintaining the
efficiency scores shown in Table 4 were significantly dif- integrity of the relationships of the data. The final data
ferent than those in Table 3. For example, the Nitric set that will be used for the analysis are shown in
Acid Recovery and Recycle technology decreased in Table 5.
efficiency when going from the first scaled data set to This section on scaling of data for the DEA models
the second set. Other technologies (e.g. Distillation and is meant as an instructional point for those researchers
Separation of Actinides from Waste Salts-1) increased and practitioners who wish to apply these models. We
in efficiency when going from the first scaled data set
shall now discuss the results of the various models using
to the second set.
the data set in Table 5.
This initial finding represents an important caveat for
scaling of data using arbitrary absolute values. Two steps
were taken to make sure that data scaling errors did not

Table 3
Scaled data set 1. Data is arbitrarily scaled by subtracting cost savings from a value of 10,000 for capital and operational costs, and adding a value
of 20 to the transuranic and low level waste values

Technology Capital costs Operational costs Transuranic waste Low level waste Simple technical
efficiency (CCR)

Nitric acid recovery 9760 9872 12 20 0.232


Distill and separate 10325 8994 10 79 0.881
actinides/salts 1
Distill and separate 11400 8983 37 46 0.564
actinides/salts 2
Waste accept. for non- 10363 9500 10 95 1.000
rad. disp.
Decon. gloveboxes 10055 9744 87 86 1.000
Chloride waste stream 5550 10246 22 6 0.458
polishing
Magnetic separation 3480 9921 7 6 0.232
Freeze drying technology 6100 9659 30 43 0.805
J. Sarkis, J. Weinrach / Journal of Cleaner Production 9 (2001) 417–427 423

Table 4
Scaled data set 2. Data is arbitrarily scaled by subtracting cost savings from a value of 7000 for capital costs and a value of 1500 from operational
costs, and adding a value of 4 to the original transuranic waste value and adding a value of 15 to the original low level waste values

Technology Capital costs Operational costs Transuranic waste Low level waste Simple technical
efficiency (CCR)

Nitric acid recovery 6760 1372 6 15 0.184


Distill and separate
7325 494 4 74 1.000
actinides/salts 1
Distill and separate
8400 483 31 41 1.000
actinides/salts 2
Waste accept. for non-
7363 1000 4 90 1.000
rad. disp.
Decon. gloveboxes 7055 1244 81 81 1.000
Chloride waste stream
2550 1746 16 1 0.546
polishing
Magnetic separation 480 1421 1 1 0.181
Freeze drying
3100 1159 24 38 1.000
technology

Table 5
Scaled data set 3. Data is scaled by actual absolute values and ratio scaled

Technology Capital costs Operational costs Transuranic waste Low level waste

Nitric acid recovery 7260 4372 3.597 0.7143


Distill and separate actinides/salts 1 7825 3494 3.571 0.7457
Distill and separate actinides/salts 2 8900 3483 3.953 0.7278
Waste accept. for non-rad. disp. 7863 4000 3.571 0.7547
Decon. gloveboxes 7555 4244 4.926 0.7496
Chloride waste stream polishing 3050 4746 3.731 0.7072
Magnetic separation 980 4421 3.534 0.7072
Freeze drying technology 3600 4159 3.846 0.7262

7. DEA models results and discussion best on the selected input and output factors is difficult
to determine. This difficulty arises from the number of
The efficiency score results of the various DEA mod- technically efficient technology (those with a score of
els and runs are shown in Table 6. The first column of 1.000). In this case five of the technologies had
results represent the simple technical efficiency scores efficiency scores of 1. Alone, the data in this first column
from the basic CCR DEA model. In this column we see may not be useful. These results can be useful if
that the determination of which technology performs additional decision models are to be used, where this

Table 6
Efficiency score results of DEA model executions

Technology Simple tech. eff. Simple X-eff Aggressive X-eff Rank efficiencies ROI

Nitric acid recovery 0.833 0.807 0.698 0.833 128%


Distill and separate 1.000 0.956 0.846 1.054 299%
actinides/salts 1
Distill and separate 1.000 0.949 0.859 1.053 63%
actinides/salts 2
Waste accept. for non-rad. 0.915 0.872 0.762 0.915 133%
disp.
Decon. gloveboxes 1.000 0.970 0.856 1.118 455%
Chloride waste stream 0.887 0.877 0.739 0.887 205%
polishing
Magnetic separation 1.000 0.989 0.890 3.112 1960%
Freeze drying technology 1.000 0.993 0.837 1.016 166%
424 J. Sarkis, J. Weinrach / Journal of Cleaner Production 9 (2001) 417–427

simple technical efficiency data is used to filter out inef- there is a strong ranking correlation between the rank
ficient technology (see [11], who applied multi-attribute efficiency scores and ROI results. Yet, Distillation and
utility theory to the robot selection process after using Separation of Actinides from Waste Salts-2 still per-
the basic DEA formulation to filter out inefficient units). forms better than other units that had higher ROI scores.
The second column of results shows the simple mean Based on these results, if there were only one tech-
cross-efficiency scores. These values were determined nology to be selected, the clear favorite would be the
by using the optimal weights from the basic CCR model. High-Gradient Magnetic Separation technology, since it
There are no clear “false positive” candidates in these performed well in terms of financial and environmental
measures. That is, those units that were initially techni- efficiency. If there were enough capital budgeted to
cally efficient still performed better then those that were allow for more than one choice then the ranking of the
technically inefficient, when the cross efficiency scores alternatives would prove more useful. A useful technique
in the second column are used to evaluate the tech- to help analyze the selection of a portfolio of alternatives
nology. The Application of Freeze-Drying Technology under a capital rationing scheme could be modeled using
to Decontamination of Liquid Wastes (0.993) and High- the efficiency scores in a mathematical programming
Gradient Magnetic Separation (0.989) technology alter- formulation. This extension is beyond the scope of the
natives had the highest simple mean cross-efficiency presentation of this paper. A difficulty arises in
scores. When compared to the ROI (in the last column determining which ranking would be the most acceptable
of Table 6), we see that 166% ROI for the highest ranked to management. An average ranking could be determined
alternative (Application of Freeze-Drying Technology to from among the various techniques, but it is difficult to
Decontamination of Liquid Wastes) is only the fifth best determine the importance assigned to one ranking
ROI. The relationships with ROI and the other approach versus another. This approach would probably
approaches will be further detailed below. require a weighting of how management feels about each
The next column in Table 6 is the aggressive mean of the ranking approaches.
cross-efficiency measure as determined by the optimal The ranking approaches, as we have mentioned, may
weights generated by the Doyle and Green formulation. contain some similarities in the order of the ranked tech-
Again, there are no obvious “false positives” that occur nology. Table 7 shows a full picture of the relationships
within this column of results. Yet, the ordering of the of the rankings based on the Best to Worst ROI ranks.
technology alternatives does change slightly. For Looking at these data we see that the rankings do not
example the High-Gradient Magnetic Separation alterna- seem to be consistent between ROI and the simple
tive has the highest mean cross-efficiency score (0.890). efficiency or either of the cross-efficiency scores. Yet,
The Application of Freeze-Drying Technology to there does seem to be a general relationship between the
Decontamination of Liquid Wastes technology which rank efficiency scores and the ROI results. To test how
was ranked highest under the simple mean cross well the various approaches correlate with each other
efficiency score is now the fifth best alternative. The Dis- a Pearson product moment correlation coefficient was
tillation and Separation of Actinides from Waste Salts- determined for each pair of rankings as shown in Table
alternative, which has the lowest ROI score (63%), 8. Table 8 results show that there is a strong correlation
ranked as the second best technological alternative at in the rankings of the ROI and ranked efficiency
0.859. approaches (0.987). And a strong correlation between
The fourth column in Table 6 shows the rank the simple technical efficiency, simple mean cross-
efficiency scores. High-Gradient Magnetic Separation is efficiency, and aggressive mean cross-efficiency scores,
the most clearly favored technology using the rank which are all above a 0.95 correlation score.
efficiency approach. As we shall see later in Table 7 These correlations may simplify the analysis slightly,

Table 7
Relationships of ROI and various DEA efficiency scores

ROI ranked measures


ROI Simple tech. eff. Simple X-eff Aggressive X-eff Rank efficiencies

1960% 1.000 0.989 0.890 3.112


455% 1.000 0.97 0.856 1.118
299% 1.000 0.956 0.846 1.054
205% 0.887 0.877 0.739 0.887
166% 1.000 0.993 0.837 1.016
133% 0.915 0.872 0.762 0.915
128% 0.833 0.807 0.698 0.833
63% 1.000 0.949 0.859 1.052
J. Sarkis, J. Weinrach / Journal of Cleaner Production 9 (2001) 417–427 425

Table 8
Pearson rank correlations for various evaluation techniques

Evaluation techniquea ROI STE SXE AXE RCCR

ROI 1.000 0.338 0.441 0.521 0.987


STE 1.000 0.967 0.972 0.390
SXE 1.000 0.951 0.481
AXE 1.000 0.573
RCCR 1.000

a
ROI=Return on Investment; STE=Simple Technical Efficiency; SXE=Simple Cross Efficiency; AXE=Aggressive Cross Efficiency;
RCCR=Rank Efficiency.

where management may want to only weigh the rankings understanding of the approach is relatively low, prim-
of each of the groups that are correlated. Yet, these cor- arily due to its mathematical complexity and use of
relations are only true for this data set, and additional mathematical programming approaches to arrive at an
data may provide significantly different correlations answer. Thus, an expert analyst would be required to aid
among the various approaches presented here. decision makers in applying DEA. Eventually, effective
and user friendly support systems may allow for it to be
more broadly accepted. DEA’s data requirements, in
8. DEA and other multiple criteria decision models terms of actual managerial inputs are not as extensive
as many of the other approaches. The lack of a require-
A number of other tools and multiple criteria decision ment for managerial preference input for DEA is advan-
models may be used for evaluation of technology. tageous since requiring management effort and time can
Included among these other tools are the analytical hier- be costly. The results of the approach are relatively easy
archy process (AHP), multiattribute utility theory to understand, because it gives a quick evaluation among
(MAUT), expert systems, multiple objective mathemat- the portfolio of technologies and is capable of ranking
ical programming such as goal programming, out- the technology based on efficiencies. Even though a var-
ranking, simulation, and scoring models. Table 9 pro- iety of parameters and factors can be included in a DEA
vides some comparative information on technique analysis, categorical and qualitative factors are still dif-
characteristics, as well as some representative references ficult to integrate.
of their applications to justification of technologies. Overall, it is difficult to compare directly the many
Much of the literature in this area focused on manufac- types of tools for evaluation of technology, each has its
turing and information technology. Clearly, these tech- disadvantages and advantages. Many times it is depen-
niques are also applicable to environmental technologies, dent on the decision environment where time, data avail-
most of which are strategically focused. ability, and costs all come into play when deciding on
In Table 9, we see that DEA does not dominate the appropriate tool.
(perform better than all other methods) in a number of As we have seen, the results from DEA are very sensi-
characteristics. From a managerial point of view it is tive to the data and the types of scaling that is used. This
moderately costly to implement such an approach from data sensitivity is a major drawback in using DEA as a
a training and data gathering context. The managerial decision tool, analysts, researchers and decision makers

Table 9
Summary of multiple criteria evaluation technique characteristics and exemplary referencesa

Evaluation technique Cost of Data Ease of Economic Management Mathematical Parameter References
implementation requirements sensitivity rigor understanding complexity mixing-
flexibility

DEA M M L M L H M [11,12]
AHP M M L L M L H [13–15]
Expert systems H H L H M H H [16,17]
Goal program M M M H L H L [18,19]
MAUT H H M M M M H [20,21]
Outranking M M L M L M M [22]
Simulation H H H H H H M [23,24]
Scoring models L L L L H L H [25,26]

a
H=High; M=Medium; L=Low (Adapted from [27]).
426 J. Sarkis, J. Weinrach / Journal of Cleaner Production 9 (2001) 417–427

should be aware of these limitations, as they should for on financial information, DEA is a multifactor ratio
any tool. For a more complete evaluation of how these approach that can incorporate a variety of factors.
and other multiple criteria evaluation techniques com- For future work, additional extensions to DEA that
pare and their underlying philosophies please see [28– could be considered in the evaluation of environmentally
30]. conscious technologies is inclusion of weight restrictions
on the input and output weights (see [31,32]). Weight
restrictions will require the preference structures of man-
9. Summary and conclusion agers to help determine the upper and lower bound con-
straints for each of the weights. This requirement means
This paper provides an additional set of techniques to that managers and decision makers would need to be
help evaluate investments in environmentally conscious more actively involved in providing importance levels
organizational technology. The use of DEA is presented for the various input and output factors. This is advan-
and evaluated from the perspective of evaluating the tageous in that it allows for more realistic weighting
environmental and financial efficiency using a multi-fac- schemes, but disadvantageous in that management must
tor productivity measurement. The small example spend time determining the importance levels of these
presented here uses actual data to evaluate environmen- factors.
tally conscious waste treatment technology at a govern- Unfortunately, the DEA approach was not available
mentally supported national research laboratory. The at the time this decision was made. From a practical per-
example showed the various types of DEA analyses that spective, managerial acceptance of these evaluations
could be used to evaluate these technologies. The needs to be determined. But, we have illustrated that
example as presented was based on two input and two these techniques may provide a viable set of tools for
output factors. This limitation was due to the number of evaluation of environmentally conscious waste treatment
alternatives that were to be considered and the technology and other technologies within an organiza-
input/output data that was available. More complex tion.
decision frameworks could be modeled as well.
Variations of the DEA analysis technique included the
use of cross-efficiency and ranking techniques. We Acknowledgements
showed that in applying DEA, managers and researchers
should be aware of the sensitivity of DEA to data and This work was partially supported by NSF Grant
data scaling. Inappropriate scaling may provide very dif- 9320949 and Texas Higher Education Coordinating
ferent answers resulting in misrepresentations of the Board ATP Grant Number 003656-036.
most appropriate technology path to pursue. Researchers
who wish to apply DEA should be aware of these limi-
tations, especially if scaling is required to adjust the data References
for DEA execution. We found that DEA is a good tool
to discriminate among the technologies and can also be [1] Camanho ASR, Dyson RG. Efficiency, size, benchmarks and tar-
used to rank the technologies if a portfolio of techno- gets for bank branches: an application of data envelopment analy-
logies were to be selected based upon a definite budget. sis. J Operational Res Soc 1999;50(9):903–15.
[2] Oral M, Kettani O, Lang P. A methodology for collective evalu-
The selection of the appropriate DEA model is not ation and selection of industrial R&D projects. Managmnt Sci
always clear. Managers should determine what their 1991;37(7):871–85.
goals are when seeking to select one of the tools. For [3] Sueyoshi T. Measuring efficiencies and returns to scale of Nippon
example, the basic formulation may be adequate if all Telegraph and Telephone in production and cost analysis. Man-
they need is to filter the “best” performers from the worst agmnt Sci 1997;43(6):779–96.
[4] Glass JC, McKillop DG, O’Rourke G. A cost indirect evaluation
performers. If a manager wishes to discriminate among of productivity change in UK universities. J Prod Anal
all the technologies applying a cross-efficiency or rank- 1999;10(2):153–75.
ing efficiency approach may be more appropriate. [5] Charnes C, Cooper WW, Lewin AY, Seiford LM. Data envelop-
In this situation we also found that DEA and ROI ment analysis: theory, methodology and application. Boston:
could be related, but it may depend on the data used and Kluwer Academic Publishers, 1993.
[6] Boerigter ST. Cost/benefit analysis for waste minimization tech-
the DEA model that is applied. For example, if cost data nologies at TA-55. Technical Report LA-UR-95-1977, Los
were to be used in both ROI and the DEA models then Alamos National Laboratory, Los Alamos, New Mexico 1995.
we can expect a significant relationship. If non-cost, [7] Charnes C, Cooper WW, Rhodes E. Measuring the efficiency of
intangible or performance data were introduced into the decision making units. Eur J Operational Res 1978;2:429–44.
models, we could expect the DEA results and ROI to be [8] Andersen P, Petersen NC. A procedure for ranking efficient units
in data envelopment analysis. Managmnt Sci
significantly different. Managers should be aware of the 1993;39(10):1261–4.
differences and assumptions associated with the data and [9] Sexton TR, Silkman RH, Hogan AJ. Data envelopment analysis;
models. Whereas ROI is a relatively simple ratio relying critique and extensions. In: Silkman RH, editor. Measuring
J. Sarkis, J. Weinrach / Journal of Cleaner Production 9 (2001) 417–427 427

efficiency; an assessment of data envelopment analysis. San Fran- [26] Semich JW. Here’s how to quantify IT investment benefits. Dat-
cisco: Jossey-Bass, 1986:73–104. amation 1994;40(1):45–7.
[10] Doyle J, Green R. Efficiency and cross-efficiency in DEA: deri- [27] Sarkis J, Sundarraj RP. Factors for strategic evaluation of
vations, meanings and uses. J Operational Res Soc enterprise Information technologies. Int J Physical Distrib Logis-
1994;45(5):567–78. tics Managmnt 2000;30(3/4):196–220.
[11] Khouja M. The use of data envelopment analysis for technology [28] Canada JR, Sullivan WG. Economic and multiattribute evaluation
selection. Comp and Indust Engng 1995;28(2):123–32. of advanced manufacturing systems. Englewood Cliffs, NJ: Pren-
[12] Sarkis J. Evaluating flexible manufacturing systems alternatives tice-Hall, 1989.
using data envelopment analysis. The Engng Economist [29] Olson DL. Decision aids for selection problems. New York:
1997;43(1):25–48. Springer, 1996.
[13] Albayrakoglu MM. Justification of new manufacturing tech- [30] Zionts S. Multiple criteria decision making; an overview and sev-
nology: a strategic approach using the analytical hierarchy pro- eral approaches. Working Paper 454, SUNY at Buffalo School
cess. Product Inventory Managmnt J 1997;37(1):71–6. of Management, Buffalo, NY 1980.
[14] Kleindorfer PR, Partovi FY. Integrating manufacturing strategy [31] Thompson RG, Langemeier LN, Lee CT, Thrall RM. The role
and technology choice. Eur J Operational Res 1990;47:214–24. of multiplier bounds in efficiency analysis with application to
[15] Sarkis J. Evaluating environmentally conscious business prac- Kansas farming. J Econometrics 1990;46(1/2):93–108.
tices. Eur J Operational Res 1998;107(1):159–74. [32] Wong YH, Beasley JE. Restricting weight flexibility in data
[16] Borenstein D. IDSSFLEX: an intelligent DSS for the design and envelopment analysis. J Operational Res Soc 1990;41(9):829–35.
evaluation of flexible manufacturing systems. J Operational Res
Soc 1998;49(7):734–44. Joseph Sarkis is currently an Associate Professor in The Graduate School
[17] Padmanabhan S. A tandem expert support system as justification of Management at Clark University. He earned his PhD from the State
for a flexible manufacturing system. J Manufacturing Syst University of New York at Buffalo. His research interests include manu-
1989;8(3):195–205. facturing strategy and management, with a specific emphasis on perform-
[18] Stam A, Kuula M. Selecting a flexible manufacturing system ance management, justification issues, enterprise modeling and environ-
using multiple criteria analysis. Int J Production Res mentally conscious operations and logistics. He has published over 120
1991;29(4):803–20. articles in a number of peer reviewed academic journals and conferences.
He is a member of the American Production and Inventory Control Society
[19] Suresh NC. An extended multi-objective replacement model for (APICS), Institute for Operations Research and Management Sciences
flexible automation investments. Int J Production Res (INFORMS), the Decision Sciences Institute (DSI), and the Production
1991;29(9):1823–44. and Operations Management Society (POMS). He is also a certified pro-
[20] Chandler JS. A multiple criteria approach for evaluating infor- duction and inventory manager (CPIM).
mation systems. MIS Quarterly 1982;6(1):61–74.
[21] Pandey PC, Kengpol A. Selection of an automated inspection Jeffrey Weinrach has many years of experience in pollution prevention
system using multiattribute decision analysis. Int J Production and waste minimization (P2/WMin). He has provided leadership to the
Economics 1995;39(3):289–98. Pollution Prevention Program Office, now called the Environmental Stew-
[22] Parsaei HR, Wilhelm MR, Kolli SS. Application of outranking ardship Office, at Los Alamos National Laboratory (LANL). He has
methods to economic and financial justification of CIM systems. worked to bring innovative approaches to LANL P2/WMin activities. He
Comput Industrial Engng 1993;25(1/4):357–60. has worked with the US Department of Energy (DOE), US Environmental
[23] Primrose PL. Investment in manufacturing technology. London: Protection Agency (EPA), US Department of Commerce, US Department
Chapman and Hall, 1991. of Defense, the State of New Mexico, and the University of New Mexico
to bring ideas and projects together. Dr. Weinrach is co-Editor-in-Chief
[24] Suresh NC, Meredith JR. Justifying multimachine systems; an of the “International Journal of Environmentally Conscious Design and
integrated strategic approach. J Manufacturing Syst Manufacturing” and serves on the Editorial Board of the “Encyclopedia
1985;4(2):117–34. of Life Support Systems.” Dr. Weinrach also serves as General Chairman
[25] Nelson CA. A scoring model for flexible manufacturing systems of the “International Congress of Environmentally Conscious Design
project selection. Eur J Operational Res 1986;24:346–59. and Manufacturing.”

You might also like