Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 113
6
A Framework for Decision Making
INTRODUCTION
Chapter 5 presented four alternatives available to Bayer: (1) continuing with
the existing process, (2) adopting an alternative chemical process not involving
MIC, (3) using an alternative process for MIC production that would consume MIC
immediately and thus not require storage, and (4) reducing the volume of stored
MIC and the risks of transporting MIC from one facility within the site to another
by rearranging process equipment. Table 6.1 summarizes how these four alterna-
tives compare along several key performance features, or “attributes”, some of
which were captured in Table 5.3 and some of which are additional attributes that
could have been considered an analysis of trade-offs. An “X” indicates that the
given manufacturing approach is superior to the others along the given attribute. As
is clear from this comparison, none of the alternatives is superior to all the others
along every attribute. For example, the Institute process, while posing higher risks
due to the volume of MIC stored, generates less wastewater than the original non-
MIC process used to produce carbaryl, and also has many cost advantages
Given that no alternative clearly dominates the others, the question arises as
to what decision-making framework could be used to identify the “best” choice.
Benefit-cost analysis (BCA), as described in Chapter 5, assigns dollar values to
the different attributes, based on their expected values, adds together all the ben-
efits and costs for each alternative, then chooses the one with the highest net
benefits. While Chapter 5 illustrated that valuing uncertain attributes can be diffi -
cult in many cases, BCA has been used extensively for many years by government
agencies, and certain conventions have been adopted (such as using the value of
a statistical life to evaluate fatality risks) to deal with some common attributes.
113
OCR for page 114
114 USE AND STORAGE OF METHYL ISOCYANATE (MIC) AT BAYER CROPSCIENCE
TABLE 6.1 Multiple Attribute Analysis of Alternative Carbamate Pesticide
Production Processes
(1)—Existing Institute Process
(2)—Non-MIC Processes
(3)—Process Producing Gaseous MIC Consumed Immediately
(4)—Alternative MIC-based Process Arrangements, Reducing Onsite Storage
Manufacturing Approach
Potentially Important Attributes
External/Regulatory Pressures (1) (2) (3) (4)
Liability for harm to surrounding communities X X
Community acceptance X ?
Liability for worker injury due to MIC exposure X ?
Liability for worker injury due to dust exposure X
Wastewater disposal requirements X X
Length of time for regulatory approval X
Internal/Cost Pressures
Product purity X X
Cost and availability of chemical feedstocks X X
Capital costs X
Equipment O&M costs X
Costs of safety equipment ? ? ? ?
Process corrosivity X X
Process yield X
Availability of in-house expertise X X
Previous experience with large-scale production X
NOTE: “X” indicates that the given pesticide manufacturing approach performs better than the others
along the given attribute. “?” indicates that information available to the committee was insufficient to
evaluate the approach along the given attribute.
The complex, multi-attribute decision-making Bayer and the legacy owners
faced when modifying the MIC and carbamate-pesticide production processes is
a challenge that will be familiar to many in the chemical industry. Incorporating
risk considerations into these analyses can be particularly challenging as the final
decision may affect individuals and organizations beyond the company itself,
e.g., reducing or increasing requirements for local emergency responders. In fact,
several experts in chemical engineering have suggested that the lack of effective
methods for analyzing such trade-offs is a major barrier to the widespread use of
inherently safer processes (ISP; Khan and Amyotte, 2004). With this in mind, the
committee chose to present one possible approach for analyzing these and other
trade-offs, while recognizing other methods are also possible.
OCR for page 115
115
A FRAMEWORK FOR DECISION MAKING
The alternative to BCA described in this chapter is multi-attribute utility
theory (MAUT), one of a suite of methods—or decision aids—developed to assist
with complex decision-making involving multiple stakeholders. MAU (described
in more detail in the next section) keeps the attributes separate throughout the
process and emphasizes the multi-dimensional nature of the decision process. The
method recognizes explicitly that different people could assign different values
to the different attributes—both in how strongly the person weights different
attributes when making the final decision and possibly even in whether the attri -
bute is considered to be positive or negative. By forcing the user to confront the
multiple attributes and decide their relative importance, MAU may help clarify a
difficult decision (or at least the difficult trade-offs involved). However, MAU is
still a relatively unfamiliar process for most companies, which is a disadvantage
when it comes to applying the method in real-world cases.
MULTI-ATTRIBUTE UTILITY THEORY
The need for trade-offs among conflicting objectives under uncertainty is
pervasive in many decisions faced by businesses, government agencies, and other
organizations. Decision scientists have conceived formal methods for balancing
such trade-offs and have established mathematically that these methods yield
the choice that maximizes the utility to the decision maker. Many approaches
to multiple-criteria decision making are available from the fields of economics
and decision analysis, including, for example, analytic hierarchy process (AHP)
analysis and fuzzy set theory. Overviews of methods for multi-attribute or crite -
ria decision analysis can be found in Belton and Stewart (2002), Figueira et al.
(2005), and DeBrucker et al. (2012). There are strengths and weaknesses for each
methodology, and these should be evaluated and understood by any user prior
to application. MAU, one example of these analysis methods, assigns a numeric
value (utility) representing a specific decision maker’s preferences to each of the
(multi-attribute) outcomes of each choice under consideration. The discussion
below illustrates how MAU could be employed in supporting decision making in
the chemical manufacturing industry, while recognizing that the AHP, fuzzy set
theory, and other decision-making frameworks also could be considered.
MAU is not a new idea to the chemical community. In 1995, the Center for
Chemical Process Safety (CCPS) published a book that suggested that MAU and
other decision aids could be used to support process safety assessments (CCPS,
1995). However, these decision aids, although employed regularly in other busi -
nesses, have yet to take hold in the chemical process industry. The CCPS’s obser-
vation from its 1995 book remains true today: “Decision aids have been applied
only to a very limited extent in risk decision problems in [the chemical process]
industry.” The CCPS indicates key obstacles to adopting these tools include lack
of familiarity with the tools among chemical process industry decision makers
and fear that the methods are either too simple or too costly. Nonetheless, the
OCR for page 116
116 USE AND STORAGE OF METHYL ISOCYANATE (MIC) AT BAYER CROPSCIENCE
committee believes that MAU and/or other techniques from decision analysis
could prove highly valuable for strengthening the integration of safety concerns
into decision making in the chemical process industry. Use of these techniques
could benefit not only those at risk due to safety breaches but the industries
themselves, as the techniques can lead to the identification of profitable safety
solutions that otherwise may have been overlooked. These tools could also assist
in strengthening the relationship between companies and communities by provid-
ing a framework for requesting and receiving input from external stakeholders.
Such input may help identify overlooked concerns or areas where additional com-
munication and outreach could be beneficial for maintaining a safe environment
and a positive relationship with the external stakeholders.
This section first provides an overview of MAU theory and its use for analyz-
ing trade-offs in complex decision problems under uncertainty. Then, it describes
some limitations of the inherent safety indexes currently being used. Finally, it
suggests an approach for employing MAU concepts in ISP assessments focused
on improving the choice of chemical manufacturing process. However, as noted
previously, MAU theory is not the sole method of approaching these difficult
questions, although it is hoped that the discussion here will demonstrate the
potential utility of these types of decision aids.
Decision Sciences and MAU Models: Background
The field of decision sciences emerged from the axioms of rational choice first
posed by mathematician John von Neumann and economist Oskar Morgenstern
in 1947. Von Neumann and Morgenstern proved that if an agent (decision maker)
has preferences with four specific characteristics (completeness, transitivity, con -
tinuity, and independence), then there must exist a mathematical equation known
as a utility function such that the decision makers’ preferences can be captured
by maximizing the equation’s expected value (von Neumann and Morgenstern,
1947; French, 1986). The utility function is usually designated as U(x), where
x=(x1, . . ., xn) is a vector representing how well a particular decision option
satisfies each of n attributes important to the decision maker. Because the utility
function includes multiple attributes, it provides a framework to consider trade-
offs among those attributes (for example, among cost, risk, and performance).
Since von Neumann and Morgenstern published the axioms of rational util-
ity, decision scientists have developed systematic methods for characterizing
utility functions (see, e.g., Keeney and Raiffa, 1976; French, 1986; Raiffa et al.,
2002). Many different mathematical forms for utility functions have been con -
ceived. Each functional form makes certain assumptions about independence of
and/or interactions among the conflicting objectives. For example, the simplest
type of utility function is linear in all the attributes:
OCR for page 117
117
A FRAMEWORK FOR DECISION MAKING
n
U ( x1 ,…, x n ) = ∑ kiU i ( xi ) (6.1)
i =1
In this function, the Ui(xi) represents individual utility functions for each of the
decision maker’s n objectives, and the ki represents weights assigned to the differ-
ent objectives. These weights reflect the value to the decision maker of an option
that offers the best possible outcome along objective i, while setting all other
objectives at their worst possible values. For this type of utility function to accu -
rately characterize preferences, certain strict independence conditions must hold
(for details, see Clemen and Reilly, 2001, and other texts on decision theory).
In essence, the decision makers’ preferences for one attribute cannot change as
levels of some other attribute change, even if the outcomes along attributes are
uncertain. Another type of utility function, requiring weaker independence condi-
tions, is the multiplicative function, expressed as follows:
1 + kU ( x1 ,…, x n ) = ∏ [kkiU i ( xi ) + 1] (6.2)
Several textbooks explain in detail the methods for determining which func -
tional form is appropriate for the decision situation at hand (e.g., Keeney and
Raiffa, 1976; Bunn, 1984; Clemen and Reilly, 2001). Formal courses in decision
analytic methods are offered in many business and engineering schools. In gen -
eral, the methods involve the following steps:
• Identifying the fundamental objectives of the decision makers and attri-
butes that can be used to represent progress along each objective;
• Eliciting individual utility functions for each attribute (the Ui(xi) in the
above equations)—functions that may be linear, concave, or convex, depending
on the decision makers’ risk tolerance;
• Testing the attributes for independence to determine the appropriate func-
tional form (e.g., linear, multiplicative, linear-multiplicative) for the MAU model;
and
• Eliciting the scaling constants (the k and ki in the above equations) for the
multi-attribute function.
The resulting MAU model will be specific to the decision makers upon
whose values it is based, and as a practical matter, it may be difficult for the
decision makers to identify their preferred attributes. Adding to the complexity,
even if different interest groups can agree on the attributes that should be con-
sidered in a decision, they may prefer different trade-offs among the attributes
(e.g., willingness to trade cost savings for decreased health hazard) or have
difficulty assigning values to the trade-offs, and they may exhibit different risk
tolerances (willingness to gamble on outcomes with the potential for high payoff
OCR for page 118
118 USE AND STORAGE OF METHYL ISOCYANATE (MIC) AT BAYER CROPSCIENCE
but also high risks of loss). As a result, different utility functions (with different
single-attribute utility functions and scaling constants) are needed to reflect the
preferences of groups with different values. It is worth noting that some efforts
to develop indexes for ISP assessments assume that there is a single correct pro -
cess design from an inherent safety perspective. However, groups with disparate
values may have differing utility functions that may lead to differences in the
preferred manufacturing process.
Although multiple MAU models may be needed to reflect different groups’
values, these models can be extremely useful in guiding negotiations among
groups in conflict (Raiffa et al., 2002). As Clemen and Reilly (2001) point out,
“Understanding trade-offs can be crucial for making progress in negotiation set -
tings.” By making values explicit, MAU models can reveal similarities and differ-
ences in the value structures of groups. As an example, MAU models can be used
to identify trade-offs between risk reductions and cost reductions. Keeney and
McDaniels (1992) show that a MAU function developed to inform strategic deci -
sions of the British Columbia Hydro and Power Authority (BC Hydro) implies
that the company decision makers value a hectare of wilderness lost at $2,500
(see Box 6.1). That is, at least $2.5 million in economic benefits would be needed
to justify a company choice that would damage a thousand hectares of wilderness.
Disagreements on certain features of a decision do not always result in
different preferred alternatives. That is, a MAU process may reveal that groups
in conflict might make similar choices in spite of their different values. When
groups do differ in their preferred alternatives, a MAU model can highlight the
main features of the decision about which groups disagree, and this may facilitate
compromise. Raiffa et al. (2002) provides detailed guidance on the use of MAU
models in negotiations among parties in conflict.
Decision analytic methods are now widely employed in business and other
applications, from fire department operations planning to nuclear power facility
siting (Keeney and Raiffa, 1976; von Winterfeldt and Edwards, 1986; Keefer et
al., 2004). The Decision Analysis Society, a specialty group within INFORMS
(Institute for Operations Research and the Management Sciences), organizes
regular conferences to exchange both academic and practical information on
decision analysis and publishes a journal, entitled Decision Analysis. Keeney and
Raiffa (1976) and von Winterfeldt and Edwards (1986), along with more recent
issues of Decision Analysis, provide many more practical examples of applica-
tions of MAU theory to decision making. The example provided in Box 6.2
shows that firms can “discover” new information about their preferences by using
MAU—in this case, an alternative that was preferred in four out of five attributes
turned out not to be the optimal choice, because of the high value the firm placed
on the fifth attribute.
Note that MAU provides a framework for making rational decisions; it does
not necessarily describe how people actually make decisions. In fact, substantial
research has indicated that individuals often do not act in accordance with the
OCR for page 119
119
A FRAMEWORK FOR DECISION MAKING
BOX 6.1
MAU Model Informs Strategic Decisions in the Hydropower Industry
The British Columbia Hydro and Power Authority (BC Hydro) has
used MAU to inform strategic decisions since the late 1980s, when the
company commissioned Ralph Keeney and Tim McDaniels to support
a comprehensive reassessment of the company’s planning processes.
Keeney and McDaniels constructed a MAU model that BC Hydro then
used to support decisions related to capital equipment upgrades, supply
planning, and other corporate strategic issues. Keeney and McDaniels
(1992) and Clemen and Reilly (2001) summarize how the MAU model
was developed and the values revealed by the model.
First, Keeney and McDaniels interviewed BC Hydro’s key decision
makers to identify their objectives and sets of attributes for measuring
progress toward those objectives. The result was the set of six funda
mental objectives and 22 attributes for measuring performance shown
in Table 6.2. Next, Keeney and McDaniels assessed a MAU function for
combining all of these attributes into a summary measure of utility for the
company. The result was a combination linearmultiplicative utility func
tion (see Keeney and McDaniels, 1992). This function then was pro
grammed into a spreadsheet to allow BC Hydro’s managers to assess
the overall utility of various decision alternatives.
Keeney and McDaniels used the resulting MAU function to illustrate
the dollar value, from the company’s perspective, of various attributes.
For example, they showed that the value of a hectare of wilderness, from
the company’s perspective, is equivalent to $2,500, so that any process
change that would cause such a loss but produce less than $2,500 in
expected gain would not be worthwhile. Similarly, they showed that two
power outages per year of 2 hours duration each to 20,000 large (com
mercial and industrial) customers was equivalent to $83 million. Keeney
and McDaniels observed that “if BC Hydro had opportunities to reduce
expected outages of that nature at a cost less than $83 million, those
opportunities would be good investments from the utility’s perspective.”
After Keeney and McDaniels’ MAU decision support model was in
place, BC Hydro’s Director of Strategic Planning commented,
The structured set of objectives has influenced BC Hydro planning in many
contexts. Two examples include our work to develop a decision framework
for supply planning, and a case study of an investment to upgrade reliability.
. . . Less obvious has been an evolution in how key senior planners view
planning issues. The notion of a utility function over a range of objectives
(rather than a single objective, like costs) is evident in many planning con
texts. The specific tradeoffs in the elicitation process are less important
than the understanding that tradeoffs are unavoidable in electricity utility
decisions and that explicit, wellstructured, informed tradeoffs can be
highly useful. (Keeney and McDaniels, 1992, p. 109, as quoted in Clemen
and Reilly, 2001)
OCR for page 120
120 USE AND STORAGE OF METHYL ISOCYANATE (MIC) AT BAYER CROPSCIENCE
TABLE 6.2 Fundamental Objectives and Attributes for Measuring Progress
along Objectives Described in Hydropower Case Study in Box 6.1
1. Maximize contribution to economic development
11. Minimize cost of electricity use (mills per kilowatt-hour in 1989 Canadian dollars)
12. Maximize funds transferred to government (annualized dividend payable)
13. Minimize economic implications of resource losses (cost of resource losses in 1989
Canadian dollars)
2. Act consistently with the public’s environmental values
21. About local environmental impacts
211. To flora (hectares of mature forest lost)
212. To fauna (hectares of wildlife habitat lost of Spatzizi Plateau quality)
213. To wildlife ecosystems hectares of wilderness lost of the Stikine Valley quality)
214. To limit recreational use (hectares of high quality recreational land lost)
215. To aesthetics (annual person-years viewing high voltage transmission lines in
quality terrain)
22. About global impacts (generation capacity in megawatts that results in “fossil fuel”
pollution)
3. Minimize detrimental health and safety impacts
31. To the public
311. Mortality (public person-years of life lost)
312. Morbidity (public person-years of disability equal in severity to that causing
employee lost work time)
32. To employees
311. Mortality (employee person-years of life lost)
312. Morbidity (employee person-years of lost work time)
4. Promote equitable business arrangements
41. Equitable pricing to different customers (constructed scale, see text)
42. Equitable compensation for concentrated local impacts (number of individuals that feel
they are inequitably treated)
5. Maximize quality of service
51. To small customers
511. Minimize outages (expected number of annual outages to a small customer annually)
512. Minimize duration of outages (average hours of outage per outage to small customers)
52. To large customers
511. Minimize outages (expected number of annual outages to a large customer annually)
512. Minimize duration of outages (average hours of outage per outage to large customers)
53. Improve new service (elapsed time until new service is installed)
54. Improve response to telephone inquiries (time until human answers the telephone)
6. Be recognized as public service oriented (constructed scale, see text)
SOURCE: Keeney and McDaniels, 1992.
OCR for page 121
121
A FRAMEWORK FOR DECISION MAKING
BOX 6.2
MAU Informs Materials Choices in the Automotive Industry
As an example of practical application of MAU to inform system de
sign in industry, consider the example of choosing materials for automo
bile frames and skins presented in Thurston (1990). Thurston designed
a MAU model for use by a French automobile manufacturer working on
longrange plans for its future vehicle fleets (those to be manufactured
in the next 510 years). The company’s design engineers faced three
options for vehicle frameandskin systems:
1. Traditional steel unibody,
2. Internal steel frame with nonstructural external polymer composite
skin, and
3. Steel and polymer composite frame with polymer composite skin.
The design engineers were perplexed about which material choice
to pursue, because although steel could be produced with the lowest
operating cost, it did not perform as well along other attributes (including
durability and flexibility) as the other choices. Thurston, in describing this
case study, noted, “When decision makers are faced with several alterna
tive systems, each system may be represented as a bundle of seemingly
incommensurate attributes. The best choice is not always clear.” In this
case, the automotive company, with support from decision analysts, used
a MAU model to help decide which of the three vehicle frameandskin
systems maximized the utility to the company, given the need to tradeoff
cost and performance attributes.
Thurston interviewed the decision makers at the company (in this
case, engineering materials design managers focused on longterm fleet
planning) to determine which attributes were important in their decision.
The decision makers identified five attributes:
1. Capital cost (billions of francs),
2. Operating cost (francs per vehicle),
3. Weight (kg),
4. Corrosion resistance (years of resistance to corrosion), and
5. Design flexibility (number of body styles possible per platform).
With the company’s engineers, Thurston then measured the perfor
mance of each of the three materials options along these five attributes.
Table 6.3 shows the results.
continued
OCR for page 122
122 USE AND STORAGE OF METHYL ISOCYANATE (MIC) AT BAYER CROPSCIENCE
BOX 6.2 Continued
TABLE 6.3 Alternative Automotive Frame-and-Skin Systems and
Their Performance Along Key Attributes
Design
Operating Flexibility
Capital Cost (Number
Cost (Francs of Bodies Corrosion
(Billion per Weight per Resistance
Design Francs) Vehicle) (Kilos) Platform) (Years)
Steel Uni-body 3 30 500 1 5
Steel Frame; PC Skin 2 40 425 5 15
Steel and PC Frame; PC Skin 1.5 45 350 5 15
SOURCE: Reproduced from Thurston, 1990.
Next, through inperson and electronic surveys of the decision makers,
Thurston determined an appropriate form for the MAU function—in this
case, multiplicative:
n
KU ( x ) + 1 = ∏ (Kk iU i ( x i ) + 1)
i =1
Thurston then assessed the individual utility functions Ui(xi) for each
of the five attributes, again through surveys of the decision makers, and
the scaling constants ki for each utility function, as well as the overall
scaling constant K. As an example, Figure 6.1 shows singleattribute
utility functions for operating cost and flexibility in number of body types
possible per platform.
As shown, these singleattribute functions convert the levels of each
option along each attribute to a value between 0 and 1 that reflects the
decision makers’ tolerance of risk. For example, the concave shapes
of these utility functions show that the decision maker is risk averse
in preferences for capital costs and flexibility, gaining more from initial
incremental cost savings and design flexibility than from subsequent
increments.
Figure 6.2 shows the resulting attribute scores and rankings of the
three design alternatives. As shown, the steel frame with polymer com
posite skin option has the highest utility. This result was counterintuitive
to the decision makers, since the polymer composite skinandframe
option scored highest along four of the five attributes, including capital
costs. However, the high variable costs of this latter system outweighed
the benefits along the other four attributes. Thus if the decision makers
OCR for page 123
123
A FRAMEWORK FOR DECISION MAKING
FIGURE 6.1 Single-attribute utility functions for cost and flexibility for the auto
industry case study described in Box 6.2. Note that the dots represent points on
the utility function assessed through structured interviews with the company’s
decision makers.
SOURCE: Thurston, 1990.
continued
OCR for page 124
124 USE AND STORAGE OF METHYL ISOCYANATE (MIC) AT BAYER CROPSCIENCE
BOX 6.2 Continued
FIGURE 6.2 Attribute levels and final rankings of the alternative auto frame-
and-skin designs discussed in Box 6.2.
SOURCE: Thurston, 1990.
had chosen the option that seemed intuitively to be the right choice, this
decision would not have accurately reflected their true preferences.
tenets of rational decision theory (Kahneman et al., 1982). Decision analysis,
by providing a framework for the decision process, can help overcome the well-
known human cognitive limitations that can lead to less-than-optimal decision-
making (Clemen and Reilly, 2001). It is also important to recognize that, as with
any modeling system, the quality of the data used in the analysis is critical, and
care must be taken to ensure that the inputs collected from the various stakeholder
communities—utility values for MAU—accurately represent their views.
Limitations of Existing Inherent Safety Indexes
Chemical engineers in the field of ISP design have conceived a number of
summary indexes intended to capture the trade-offs in objectives embodied by
OCR for page 125
125
A FRAMEWORK FOR DECISION MAKING
different design options. Examples include the integrated inherent safety index
(I2SI) and a set of European Union (EU) indexes known as the INSET Toolkit.
(For descriptions of these methods, see Chapter 4 and Khan and Amyotte, 2004.)
When considered from a formal decision analysis perspective, these indexes
have a number of limitations, both theoretical and practical. The main theoreti -
cal weakness is that the indexes were not designed to follow the von Neumann-
Morgenstern model of rational choice, so there is no guarantee that the index
value for a given manufacturing process will be able to reflect a given decision
maker’s actual preferences and attitudes toward risk. Perhaps more importantly,
these indexes—in contrast to a MAU approach—do not allow for the possibility
of multiple decision makers with different preferences.
As an example, consider the I2SI index (Khan and Amyotte, 2004). This
index is intended to combine assessments of multiple inherent safety attributes
into a single numeric value. The equation for computing this index is:
ISI alt
PHCI alt
I 2SI = (6.3)
HI base
PHCI base
where ISIalt, PHCIalt, HIbase, and PHCIbase are all subindexes computed as func-
tions of various process design attributes. For each manufacturing process con-
sidered, the ratio HIbase/PHCIbase is the same, so that when comparing multiple
alternatives the above equation reduces to:
5
∑ ISI } 2
Min{200, i
ISI alt i =1
I 2SI = c × =c× (6.4)
10
PHCI alt
∑ PHCI j
j =1
where c is a constant and the summation terms in the equation represent the attri -
butes listed in Table 6.4. The values for the summation terms are determined from
a combination of subjective judgments about the degree to which each process
unit satisfies the principles of ISP.
An index such as the I2SI provides a single calculation that cannot be
adjusted to reflect the variation in preferences among attributes and willingness
to tolerate risk that different constituencies may exhibit. For example, a company
owner may be willing to tolerate a small risk of a spill that could have health
effects in the community if the alternative involved a much higher risk of a fire
that would seriously damage the facility, whereas members of the community
may not accept such a trade-off, and employees of the firm (who place some value
on keeping the facility intact in order to retain their jobs) may prefer something
in between the owner and the community. In addition to putting different weights
OCR for page 126
126 USE AND STORAGE OF METHYL ISOCYANATE (MIC) AT BAYER CROPSCIENCE
TABLE 6.4 Attributes Included in the I2SI (Equation 6.4)
Attribute Category : Adherence to principles of inherently safer design
Representation in
Attribute Description I2SI Equations
Extent to which process minimizes use of hazardous materials ISI1 = ISIm
Extent to which process substitutes safer materials for more hazardous ones ISI2 = ISIsu
Extent to which process attenuates risks by operating under safer conditions ISI3 = ISIa
(e.g., room temperature and pressure)
Extent to which process simplifies manufacturing (e.g., by avoidance of ISI4 = ISIsi
multiproduct or multiunit operations or congested pipe or unit settings)
Extent to which process limits potential negative consequences of out-of- ISI5 = ISIl
normal operations (e.g., by unit segregation)
Attribute Category : Need for add-on processes to control hazards
Representation in
Attribute Description I2SI Equations
Pressure control required PHCI1 = PHCIp
Temperature control required PHCI2 = PHCIt
Flow control required PHCI3 = PHCIf
Level control required PHCI4 = PHCIl
Concentration control necessary PHCI5 = PHCIc
Inert venting necessary PHCI6 = PHCIiv
Blast wall needed PHCI7 = PHCIbw
Fire resistance wall needed PHCI8 = PHCIfr
Sprinkler system necessary PHCI9 = PHCIs
Forced dilution needed PHCI10 = PHCId
on different attributes, a MAU model can reflect differences in risk tolerances in
the form of the utility function: linear functions represent risk neutrality; concave
functions represent a preference for gambling on high risks that have potentially
high payoffs; and convex functions represent risk aversion (for details, see Cle -
men and Reilly, 2001).
The existing indexes are the proverbial black box: input a set of numbers
based on the process being evaluated, and the index produces a single value for
each alternative, which is then used to rank the different alternatives and identify
the optimal decision. All of the trade-offs, uncertainties, and risk tolerances are
hidden from view because they are implicitly assumed in the underlying calcula -
tions, rather than explicitly chosen parameters. Because these indexes implicitly
assume one value structure, the effects of alternative value structures on the pref -
erence ordering of the alternatives cannot be assessed. Indeed, because the trade-
offs in the index are completely opaque (to the analysts as well as to others), it is
unlikely that companies will be able to use such indexes to build trust within the
communities in which their facilities are located. A more transparent approach,
OCR for page 127
127
A FRAMEWORK FOR DECISION MAKING
that is, one that makes the trade-offs in attributes and risk tolerances explicit, is
needed if the outcomes of ISP assessments are to be widely embraced.
EMPLOYING MAU MODELS IN ISP ASSESSMENTS
The existing inherent safety indexes could serve as a starting point for
building MAU functions to inform process design choices, with inherent safety
in mind. For ISP design, fundamental objectives have been described as elimi -
nation, minimization, substitution, moderation, and simplification. The indexes
provide useful starting points for constructing MAU functions in that they
identify a number of the attributes by which progress along these fundamental
objectives could be measured. For example, the attributes in the I2SI include
costs associated with any damage that might occur due to a safety breach
and costs associated with a process control option under consideration. Other
attributes that could be included might be the number of fatalities that could
occur in an accident and the potential loss of community goodwill (which can
lead to additional costs for the company in its future decision-making). CCPS
(1996), mentioned at the beginning of this chapter, provides an illustration of
the use of MAU to choose a process control device for a hypothetical chemical
distillation column.
EMPLOYING MAU MODELS AT BAYER CROPSCIENCE
MAU and ISP decision-making tools could have been used to inform manu-
facturing process design choices at numerous points in the history of the Institute
pesticide plant, starting with the introduction of MIC to the site in 1978. As noted
in Chapter 5, changes to the production process made at the Institute facility were
generally considered in response to business conditions or external pressures,
without explicit consideration of ISP principles. Several of these decision points
could have provided opportunities to introduce MAU approaches, which would
have recognized the multi-attribute nature of the decision and the differences in
preferences between the firm and the community, perhaps resulting in a decision-
making process that would have been more acceptable to the community. That,
in turn, might have allowed production to continue at the plant in some form.
Chapter 5 mentions the 1984 Bhopal accident and the 2008 methomyl acci-
dent as significant opportunities for broadening the approach to decision making
at Institute. Other opportunities arose when the plant changed ownership. Choices
that better accounted for the multiple costs and benefits involved—including the
costs associated with risks imposed on the community—could have prevented
the types of accident risks that persisted until the use of MIC ceased at the site.
Without such a multi-attribute decision framework in place, the decisions at the
Institute site seem to have focused on production costs and the business risk of
interrupting the flow of product to the market, which resulted in the decision to
OCR for page 128
128 USE AND STORAGE OF METHYL ISOCYANATE (MIC) AT BAYER CROPSCIENCE
continue with the same basic production process with some modifications rather
than adopting an entirely new approach.
This bias in favor of an existing production process is not surprising and may
even reflect the optimal decision, especially from the company’s point of view.
“Steel in the ground” is a powerful motivator. It avoids the up-front capital costs
of a new process, along with any uncertainties about how well the process will
operate or what its operating cost will be. Critically and objectively reviewing a
process that has been in operation for many years can be difficult for those work -
ing at a chemical production facility. For example, from a practical standpoint, it
is likely that, over time, a given process will have been modified from its initial
design. Minor changes in procedures, modifications to existing equipment, and
the effect of age and maintenance on the system must be taken into account dur-
ing the trade-off analysis. This may be challenging if, due to staffing changes or
insufficient documentation, the modifications from the initial design have not
been recorded. In addition, alternative processes may always seem “hypothetical,”
and concerns about risks expressed by community members may be ascribed to
their lack of understanding of the process and its many layers of safety protec -
tions. These factors make it difficult to apply MAU (and ISP) analyses to exist -
ing plants in ways that can really identify promising alternatives to the existing
process.
Implementing a structured, multi-attribute decision process such as MAU
analysis may be easier when designing a new production process or an entirely
new facility. In such cases, no incumbent process has an advantage in terms of
capital costs or production uncertainties, since all the alternative processes are
new and hence hypothetical. In addition, the company may be involved in negoti-
ations with regulators—who, in turn, are considering the concerns of community
groups—in order to obtain an operating permit, so the need to address concerns
held by those outside the facility will be more salient. During these initial inter-
actions between the facility and its neighbors, a MAU analysis could serve to
educate community members about the trade-offs between risks and economic
considerations of importance to the company, while informing the company of
the community’s concerns about different types of production risks.
However, conducting a MAU analysis for the MIC issue would not neces-
sarily have resulted in different decisions by the companies owning the Institute
plant. Especially using the company’s own valuation on the multiple attributes
of the alternatives, continuing with the existing MIC process might well have
seemed optimal. The advantage that would be gained from using multi-criteria
analysis is that the company would have explicitly identified the most important
attributes of the alternative processes and assigned a valuation to those attributes.
Using MAU could also have provided a calculation into which the preferences of
others (e.g., community members) could easily be incorporated.
OCR for page 129
129
A FRAMEWORK FOR DECISION MAKING
CONCLUSIONS
No one carbamate pesticide production process dominates all the others along
every attribute that owners of the Institute site considered or could have consid-
ered. However, given the necessary trade-offs, a decision-analysis approach, such
as MAU, could have helped the various plant owners consider all aspects of their
choices. A decision analysis approach might also have facilitated the communi-
cation process between the company and the surrounding community, allowing
the two groups to compare their preferences regarding the decision attributes
and their potentially differing tolerances and attitudes toward risk. Although the
CCPS has advocated for the development of such an approach since 1995, it has
not become institutionalized in the chemical process industry. A new decision-
making framework, incorporating some of the work done to develop exist-
ing ISP indexes but also allowing explicit consideration of differences in
decision makers’ preferences across multiple attributes, could assist in the
incorporation of ISP considerations into decision making in the chemical
manufacturing industry and communication of those considerations to a
concerned public.
Design decisions cannot be strictly objective with regards to ISP as these
choices will always require trade-offs among attributes and varying levels of risk;
different individuals or constituencies may have different value systems and thus
make different trade-offs. However, a new decision framework could support
incorporation of the attributes in existing indexes, while adhering, and drawing
benefit from, the mathematics of multi-attribute or -criteria decision analysis.
The committee recommends that the Chemical Safety Board or other
appropriate entity convene a working group to chart a plan for incorporat-
ing decision theory frameworks into ISP assessments. The working group
should include experts in chemical engineering, ISP design, decision sciences,
negotiations, and other relevant disciplines. The working group should iden -
tify obstacles to employing methods from the decision sciences in process
safety assessments. It should identify options for tailoring these methods to
the chemical process industry and incentives that would encourage their use.
REFERENCES
Belton, V.,and T. S. Stewart. 2002. Multiple Criteria Decision Analysis: An Integrated Approach.
Norwell: Klumber Academic Publishers.
Bunn, D. W. 1984. Applied Decision Analysis. New York: McGraw-Hill.
CCPS (Center for Chemical Process Safety). 1995. Tools for Making Acute Risk Decisions with
Chemical Process Safety Applications. New York: American Institute of Chemical Engineers.
Clemen, R. T., and T. Reilly. 2001. Making Hard Decisions with Decision Tools. Belmont, CA:
Duxbury Press.
DeBrucker, K., Macharis, C., and A. Verbeke. 2012. Multi-criteria analysis and the resolution of
sustainable development dilemmas: a stakeholder management approach. European Jounral of
Operational Research. Available online.
OCR for page 130
130 USE AND STORAGE OF METHYL ISOCYANATE (MIC) AT BAYER CROPSCIENCE
Figueira, J., S. Greco, and M. Ehrgott. 2005. Multiple Criteria Decision Analysis: State of the Art
Surveys. New York: Springer Science and Business Media, Inc.
French, S. 1986. Decision Theory: An Introduction to the Mathematics of Rationality. Chichester:
Ellis Horwood.
Kahneman, D., P. Slovic, and A. Tversky. 1982. Judgment Uner Uncertainty: Heuristics and Biases.
Cambridge, UK: Cambridge University Press.
Keefer, D. L., C. W. Kirkwood, and J. L. Corner. 2004. Perspective on decision analysis applications,
1990-2001. Decision Anal. 1(1):4-22.
Keeney, R., and T. McDaniels. 1992. Value-focused thinking about strategic decisions at BC. Hydro.
Interfaces 22(6):94-109.
Keeney, R., and H. Raiffa. 1976. Decisions with Multiple Objectives: Preferences and Value Tradeoffs.
New York: Wiley.
Khan, F. I., and P. R. Amyotte. 2004. Integrated Inherent Safety Index (I2SI): A tool for inherent safety
evaluation. Process Saf. Prog. 23(2):136-148.
Raiffa, H., J. Richardson, and D. Metcalfe. 2002. Negotiation Analysis: The Science and Art of Col-
laborative Decision Making. Boston: Belknap Harvard.
Thurston, D. L. 1990. Multiattribute utility analysis in design management. IEEE Trans. Eng. Man-
age. 37(4):296-301.
von Neumann, J., and O. Morgenstern. 1947. Theory of Games and Economic Behavior, 2nd Ed.
Princeton, NJ: Princeton University Press.
von Winterfeldt, D., and W. Edwards. 1986. Decision Analysis and Behavioral Research. Cambridge,
UK: Cambridge University Press.
